Wide dynamic range adjustment method and system based on vehicle image sensor

Abstract

The present invention relates to the field of image processing, and in particular to a wide dynamic range adjustment method and system based on a vehicle-mounted image sensor. The method comprises the following steps: acquiring real-time vehicle-mounted recorded images based on the vehicle-mounted image sensor; performing dynamic delay interpolation optimization and motion blur compensation on the real-time vehicle-mounted recorded images to construct distortion-optimized vehicle-mounted images; performing ambient light source recognition on the distortion-optimized vehicle-mounted images, and adjusting dynamic image spectrum parameters to generate dynamic spectrum-adjusted images; performing pixel-by-pixel histogram calculations on the dynamic spectrum-adjusted images, and performing dynamic exposure range distribution visualization to construct a dynamic exposure range distribution perception map; performing region-by-region adaptive exposure compensation calculations and multi-parameter exposure compensation adjustments based on the dynamic exposure range distribution perception map to extract optimal exposure compensation parameters. The present invention improves the image quality of vehicle-mounted images by improving the dynamic range of vehicle-mounted images, and provides clear and accurate visual information.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a wide dynamic range adjustment method and system based on a vehicle-mounted image sensor. Background Art

[0002] With the rapid development of autonomous driving technology and intelligent transportation systems, vehicle-mounted image sensors have become an indispensable component of modern vehicles. Vehicle-mounted image sensors are responsible for capturing visual information of the vehicle's surroundings and providing key data support for autonomous driving systems. However, vehicle-mounted image sensors often face strong lighting contrast, shadow areas, and highlight areas under complex lighting conditions, which poses a great challenge to image capture and processing.

[0003] When faced with these high dynamic range (HDR) scenes, traditional image sensors are often unable to retain detailed information of both bright and dark areas, resulting in overexposure or underexposure in the image, affecting the system's accurate perception of the environment. Especially in complex scenes such as high-speed driving, tunnel entrances and exits, and strong direct sunlight, traditional sensors often cannot meet the requirements of high dynamic range, greatly limiting the image perception capabilities. Therefore, in order to better adapt to the needs of modern intelligent transportation systems and autonomous driving, there is an urgent need for an efficient, real-time, and wide dynamic range adjustment method for vehicle-mounted image sensors that can adapt to various complex lighting environments. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a wide dynamic range adjustment method and system based on a vehicle-mounted image sensor to solve at least one of the above technical problems.

[0005] To achieve the above object, the present invention provides a wide dynamic range adjustment method based on a vehicle-mounted image sensor, comprising the following steps:

[0006] Step S1: acquiring a real-time vehicle-mounted recorded image based on a vehicle-mounted image sensor; and performing dynamic delay interpolation optimization and motion blur compensation on the real-time vehicle-mounted recorded image to construct a distortion-optimized vehicle-mounted image.

[0007] Step S2: identifying the ambient light source for the distortion-optimized vehicle-mounted image, and adjusting the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image;

[0008] Step S3: Calculate the histogram of each pixel of the dynamic spectrum adjustment image, and visualize the dynamic exposure range distribution to construct a dynamic exposure range distribution perception map;

[0009] Step S4: performing region-by-region adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment according to the dynamic exposure range distribution perception map, and extracting optimal exposure compensation parameters;

[0010] Step S5: accurately locate the bounding boxes of each object one by one according to the real-time vehicle-mounted recorded images, and perform dynamic movement trajectory fitting to generate the movement trajectory of each dynamic object;

[0011] Step S6: Perform dynamic tracking exposure decision according to the moving trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

[0012] The present invention uses an on-board image sensor to collect data in real time, ensuring that the on-board system can capture all relevant information in a dynamic driving environment, including objects around the vehicle, road conditions, traffic signs, etc. This provides rich real-time image materials for subsequent processing. The dynamic delay interpolation optimization technology can reduce the timing errors caused by image acquisition delays and ensure that the time difference between multiple consecutive frames is minimized. This is particularly important for vehicles traveling at high speeds, and can avoid image distortion caused by delays. In the case of high-speed driving or fast object movement, motion blur occurs in the image. By using motion estimation and deblurring algorithms (such as deblurring technology based on phase consistency), motion blur is effectively compensated, image details are restored, and image clarity and sharpness are improved. By identifying the ambient light source in the on-board image, the system can determine the color temperature, intensity and distribution of the light source, thereby dynamically adapting to different lighting conditions (such as direct sunlight, cloudy days, nighttime, etc.). This light source recognition can greatly improve the adaptability of the image processing system under complex lighting conditions. According to the characteristics of the ambient light source, the system adjusts the spectral parameters of the image such as color temperature, saturation, contrast, etc. in real time to ensure that the image color is natural and true, and remains stable under different lighting conditions. This helps avoid image color distortion or uneven lighting in scenes with strong contrast, improving visual effects and information readability. By calculating the histogram of brightness values ​​for each pixel, the brightness distribution of the entire image is accurately analyzed, and overexposed or underexposed areas in the image are discovered in a timely manner. This provides very accurate data support for subsequent dynamic exposure adjustments. A dynamic exposure range distribution perception map is constructed to intuitively display the dynamic range of the image and show the brightness level of each area. This not only helps to identify exposure problems in real time, but also provides a basis for subsequent local exposure adjustments. Based on the exposure conditions of different areas, the exposure compensation of each area is automatically calculated based on the regional characteristics (such as overexposure, underexposure, and normal exposure). This method ensures that the brightness level of each area is accurately adjusted, avoiding some areas in the image from being too bright or too dark, and ensuring overall exposure balance. Through multi-parameter adjustment (such as exposure time, ISO gain, etc.), the system finely adjusts the brightness of each area so that the details in the image are fully preserved, especially in high-contrast environments, to avoid detail loss. Extract the optimal parameter settings from multiple exposure adjustment schemes to ensure the most ideal exposure effect for the overall image, which can avoid overexposure and underexposure problems of the image and fully display the details and layering. Through high-precision object detection algorithms (such as YOLO, Faster R-CNN, etc.), the bounding box of the object can be accurately marked in the image, providing a basis for subsequent object tracking. This enables the system to accurately identify key objects such as lane lines, pedestrians, and traffic signs to avoid missed detection or false detection. For moving objects (such as other vehicles or pedestrians), object tracking algorithms (such as Kalman filtering, Deep SORT, etc.) are used to track them in real time and fit their motion trajectories.This not only helps the system better understand the behavior of objects in the scene, but also provides real-time data support for subsequent dynamic exposure control. Based on the motion trajectory of each dynamic object, the system dynamically adjusts the exposure strategy according to the speed, direction and lighting conditions of the object. For fast-moving objects (such as cars), the system quickly adjusts the exposure time, while for slower objects (such as pedestrians), a smooth exposure adjustment strategy is used. Based on real-time dynamic object tracking and optimal exposure compensation parameters, the system performs end-to-end exposure control to ensure that each dynamic object remains clear, without overexposure or underexposure during movement. This intelligent exposure control can significantly improve the dynamic range and detail performance of vehicle-mounted images, especially in high dynamic range scenes. Through multiple rounds of dynamic exposure adjustment decisions, the system continuously learns and optimizes the exposure control model to adapt to different driving environments. This adaptive dynamic exposure control model improves the adaptability of vehicle-mounted images in complex scenes, thereby ensuring the best image quality.

[0013] In this specification, a wide dynamic range adjustment system based on a vehicle-mounted image sensor is provided, which is used to perform the wide dynamic range adjustment method based on the vehicle-mounted image sensor as described above, including:

[0014] A distortion optimization module is used to obtain real-time vehicle-mounted recorded images based on the vehicle-mounted image sensor; dynamic delay interpolation optimization and motion blur compensation are performed on the real-time vehicle-mounted recorded images to construct distortion-optimized vehicle-mounted images;

[0015] The spectrum adjustment module is used to identify the ambient light source of the distortion-optimized vehicle-mounted image, and adjust the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image;

[0016] The exposure distribution perception module is used to calculate the histogram of each pixel of the dynamic spectrum adjustment image, visualize the dynamic exposure range distribution, and construct a dynamic exposure range distribution perception map;

[0017] An exposure compensation module is used to perform adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment for each area according to the dynamic exposure range distribution perception map, and extract the optimal exposure compensation parameters;

[0018] The object tracking module is used to accurately locate the bounding box of each object based on the real-time vehicle-mounted recorded images, and to fit the dynamic movement trajectory to generate the movement trajectory of each dynamic object;

[0019] The dynamic tracking exposure module is used to make dynamic tracking exposure decisions according to the movement trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

[0020] The present invention uses the on-board image sensor in scenes of high-speed driving or fast movement, and the image will be distorted due to delay. Through delayed interpolation optimization, the time dislocation and inter-frame delay of the image are reduced, the image is smooth and consistent with the timing, and the image distortion caused by time deviation is reduced. Motion blur is a common problem in high-speed or dynamic scenes. Through motion estimation and blur compensation algorithms, the blurred part is effectively restored to make the image details clearer. Motion blur compensation technology is particularly important to improve the perception of the real-time environment by the automatic driving system and ensure that key objects and road signs can remain clear even in motion. Distortion optimization and motion blur compensation work together to ensure the quality of the on-board image, and can still provide clear and accurate image data for subsequent processing even in fast driving or complex scenes. By real-time identification of the light source in the image, the system can distinguish different lighting conditions (such as daylight, cloudy days, nighttime, artificial light sources, etc.). The identification of ambient light sources helps the system understand the lighting environment of the current scene, and further adjusts the color and brightness of the image to avoid overexposure or color distortion. By adjusting the image's spectral parameters such as color temperature, saturation, and contrast, the system optimizes the image's visual effects to ensure that image details are clearly presented under different lighting environments. In particular, in strong lighting changes (such as day-to-night transitions, light reflections, etc.), it can effectively avoid overexposure or underexposure and maintain the image's natural color and layering. This dynamic spectral adjustment ensures that the image can present the best color and brightness in different environments, allowing the vehicle perception system to obtain accurate information in complex lighting environments (such as tunnels, night driving, and reflected light). The brightness distribution in the image is accurately captured by calculating the brightness value histogram of each pixel. This enables the system to identify overexposed and underexposed areas and make targeted adjustments to these areas. A dynamic exposure range distribution perception map is constructed to intuitively display the exposure status of each area in the image, helping the system to fully understand the brightness distribution of the image and discover potential exposure problems. This visualization method enables the system to more accurately determine which areas require exposure compensation during real-time driving and formulate corresponding adjustment strategies to ensure that details can be clearly presented in scenes with a wide dynamic range. The system performs adaptive exposure compensation for different areas (such as highlight areas, shadow areas, etc.) based on the exposure distribution perception map. By analyzing the exposure requirements of each area, the system intelligently adjusts parameters such as exposure time and ISO gain to ensure balanced exposure of the entire image. Not only does it adjust the exposure time, but it also combines multiple parameters such as ISO gain to perform exposure compensation, so that the brightness and details of each area in the image are optimally balanced. By dynamically compensating for each area, the phenomenon of some areas in the image being too dark or too bright is avoided, ensuring that details in high dynamic range scenes are retained and improving the visual effect of the overall image.Through high-precision object detection algorithms (such as YOLO, Faster R-CNN, etc.), the system can accurately detect and mark various objects in the image (such as pedestrians, vehicles, traffic signs, etc.). The accurate positioning of the object provides the basis for subsequent object tracking and exposure adjustment. Combined with the detection results of the object, the tracking algorithm (such as Kalman filtering, SORT, etc.) is used to fit the motion trajectory of the object in real time. This motion trajectory tracking can help the system understand the movement trend of the object and provide more accurate information for dynamic exposure control. Object tracking technology enables the system to understand the state of the dynamic object in front in real time when the vehicle is driving at high speed and respond in time, thereby improving the safety and accuracy of the autonomous driving system. Based on the motion trajectory of the object, the system can quickly adjust the exposure of fast-moving objects (such as a moving car), while using a smooth exposure strategy for slow-moving objects (such as pedestrians). This dynamic adjustment ensures that the objects in the image are always clearly visible in high-speed driving environments. By combining the dynamic trajectory of the object and the optimal exposure compensation parameters, the system controls the exposure of the entire image end-to-end to ensure that each dynamic object can adapt to different lighting conditions during movement. For example, a fast-moving car will maintain clear details under strong backlight. This precise dynamic exposure control is critical for autonomous driving systems, ensuring that the system always provides clear and accurate visual information in complex and changing lighting environments, thereby improving driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic flow chart of the steps of a wide dynamic range adjustment method based on a vehicle-mounted image sensor according to the present invention;

[0022] Figure 2 Detailed implementation flow chart of step S1;

[0023] Figure 3 Detailed implementation flow chart of step S2;

[0024] Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0025] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0026] The present application example provides a wide dynamic range adjustment method and system based on a vehicle-mounted image sensor. The execution subject of the wide dynamic range adjustment method and system based on a vehicle-mounted image sensor includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.

[0027] See also Figures 1 to 4 The present invention provides a wide dynamic range adjustment method based on a vehicle-mounted image sensor, and the wide dynamic range adjustment method based on the vehicle-mounted image sensor comprises the following steps:

[0028] Step S1: acquiring a real-time vehicle-mounted recorded image based on a vehicle-mounted image sensor; and performing dynamic delay interpolation optimization and motion blur compensation on the real-time vehicle-mounted recorded image to construct a distortion-optimized vehicle-mounted image.

[0029] Step S2: identifying the ambient light source for the distortion-optimized vehicle-mounted image, and adjusting the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image;

[0030] Step S3: Calculate the histogram of each pixel of the dynamic spectrum adjustment image, and visualize the dynamic exposure range distribution to construct a dynamic exposure range distribution perception map;

[0031] Step S4: performing region-by-region adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment according to the dynamic exposure range distribution perception map, and extracting optimal exposure compensation parameters;

[0032] Step S5: accurately locate the bounding boxes of each object one by one according to the real-time vehicle-mounted recorded images, and perform dynamic movement trajectory fitting to generate the movement trajectory of each dynamic object;

[0033] Step S6: Perform dynamic tracking exposure decision according to the moving trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

[0034] The present invention uses an on-board image sensor to collect data in real time, ensuring that the on-board system can capture all relevant information in a dynamic driving environment, including objects around the vehicle, road conditions, traffic signs, etc. This provides rich real-time image materials for subsequent processing. The dynamic delay interpolation optimization technology can reduce the timing errors caused by image acquisition delays and ensure that the time difference between multiple consecutive frames is minimized. This is particularly important for vehicles traveling at high speeds, and can avoid image distortion caused by delays. In the case of high-speed driving or fast object movement, motion blur occurs in the image. By using motion estimation and deblurring algorithms (such as deblurring technology based on phase consistency), motion blur is effectively compensated, image details are restored, and image clarity and sharpness are improved. By identifying the ambient light source in the on-board image, the system can determine the color temperature, intensity and distribution of the light source, thereby dynamically adapting to different lighting conditions (such as direct sunlight, cloudy days, nighttime, etc.). This light source recognition can greatly improve the adaptability of the image processing system under complex lighting conditions. According to the characteristics of the ambient light source, the system adjusts the spectral parameters of the image such as color temperature, saturation, contrast, etc. in real time to ensure that the image color is natural and true, and remains stable under different lighting conditions. This helps avoid image color distortion or uneven lighting in scenes with strong contrast, improving visual effects and information readability. By calculating the histogram of brightness values ​​for each pixel, the brightness distribution of the entire image is accurately analyzed, and overexposed or underexposed areas in the image are discovered in a timely manner. This provides very accurate data support for subsequent dynamic exposure adjustments. A dynamic exposure range distribution perception map is constructed to intuitively display the dynamic range of the image and show the brightness level of each area. This not only helps to identify exposure problems in real time, but also provides a basis for subsequent local exposure adjustments. Based on the exposure conditions of different areas, the exposure compensation of each area is automatically calculated based on the regional characteristics (such as overexposure, underexposure, and normal exposure). This method ensures that the brightness level of each area is accurately adjusted, avoiding some areas in the image from being too bright or too dark, and ensuring overall exposure balance. Through multi-parameter adjustment (such as exposure time, ISO gain, etc.), the system finely adjusts the brightness of each area so that the details in the image are fully preserved, especially in high-contrast environments, to avoid detail loss. Extract the optimal parameter settings from multiple exposure adjustment schemes to ensure the most ideal exposure effect for the overall image, which can avoid overexposure and underexposure problems of the image and fully display the details and layering. Through high-precision object detection algorithms (such as YOLO, Faster R-CNN, etc.), the bounding box of the object can be accurately marked in the image, providing a basis for subsequent object tracking. This enables the system to accurately identify key objects such as lane lines, pedestrians, and traffic signs to avoid missed detection or false detection. For moving objects (such as other vehicles or pedestrians), object tracking algorithms (such as Kalman filtering, Deep SORT, etc.) are used to track them in real time and fit their motion trajectories.This not only helps the system better understand the behavior of objects in the scene, but also provides real-time data support for subsequent dynamic exposure control. Based on the motion trajectory of each dynamic object, the system dynamically adjusts the exposure strategy according to the speed, direction and lighting conditions of the object. For fast-moving objects (such as cars), the system quickly adjusts the exposure time, while for slower objects (such as pedestrians), a smooth exposure adjustment strategy is used. Based on real-time dynamic object tracking and optimal exposure compensation parameters, the system performs end-to-end exposure control to ensure that each dynamic object remains clear, without overexposure or underexposure during movement. This intelligent exposure control can significantly improve the dynamic range and detail performance of vehicle-mounted images, especially in high dynamic range scenes. Through multiple rounds of dynamic exposure adjustment decisions, the system continuously learns and optimizes the exposure control model to adapt to different driving environments. This adaptive dynamic exposure control model improves the adaptability of vehicle-mounted images in complex scenes, thereby ensuring the best image quality.

[0035] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of a method for adjusting a wide dynamic range based on a vehicle-mounted image sensor according to the present invention. In this example, the method for adjusting a wide dynamic range based on a vehicle-mounted image sensor comprises:

[0036] Step S1: acquiring a real-time vehicle-mounted recorded image based on a vehicle-mounted image sensor; performing dynamic delay interpolation optimization and motion blur compensation on the real-time vehicle-mounted recorded image to construct a distortion-optimized vehicle-mounted image;

[0037] In this embodiment, a high-resolution vehicle-mounted image sensor is selected, and a CMOS sensor is usually used, which has a high dynamic range and low-light performance. The parameters of the sensor, such as the frame rate (such as 30fps) and the resolution (such as 1920x1080), are configured to ensure that clear images can be captured in different driving environments. During driving, the vehicle-mounted image is collected in real time by the sensor. The image data is processed and stored by the vehicle-mounted computer. In order to ensure the stability and accuracy of the data, it is recommended to adopt a real-time data stream processing framework, such as using ROS (Robot Operating System) or other real-time data processing tools. After the image is collected, preliminary preprocessing is performed, including denoising, white balance and color correction. The purpose of this step is to improve the image quality and ensure that better results can be obtained in subsequent processing. This process uses a Gaussian filter for denoising and adaptive histogram equalization for color correction. During the image acquisition process, the image quality is monitored in real time to ensure the integrity and accuracy of each frame of data. Set a threshold, such as the degree of motion blur does not exceed a certain value, to determine whether the image meets the requirements of subsequent processing. Select a suitable dynamic delay interpolation algorithm, such as an optical flow method or a time domain-based interpolation method. The optical flow method can estimate the intermediate frame by analyzing the motion information between consecutive frames, thereby effectively reducing the image distortion caused by delay. By calculating the optical flow field between adjacent frames, the motion information of each pixel is extracted. This step uses the Lucas-Kanade method to perform local motion estimation, obtain motion vectors, and then generate intermediate frames. Use the obtained motion information for dynamic interpolation. According to the motion vector, the adjacent frames are weighted averaged to generate new intermediate frames. Set the interpolation ratio, such as inserting one frame between every two frames, and control the smoothness of the interpolation by adjusting the scale factor. Evaluate the generated interpolated image, and use indicators such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) to evaluate the interpolation effect. Adjust the interpolation parameters according to the results to achieve the best visual effect. In the acquired real-time image, first detect the presence of motion blur. Use the Laplace operator or Fourier transform to analyze the frequency distribution of the image and evaluate the degree of image blur. Set a blur threshold. If the blur exceeds 0.2, it is identified as a motion blurred image. For the detected motion blurred image, calculate its blur degree. Obtain blur intensity by calculating the gradient value of each pixel, and determine the direction and magnitude of blur based on the intensity information. Select a suitable motion blur compensation algorithm, such as a deblurring method based on deconvolution. These algorithms can improve image quality by restoring details of blurred images. Use blind deconvolution technology combined with blur kernel estimation to perform image deconvolution. Evaluate the quality of the compensated image and use indicators such as PSNR and SSIM to evaluate the compensation effect. Adjust the compensation parameters based on the evaluation results to improve the quality of the final image.

[0038] Step S2: identifying the ambient light source for the distortion-optimized vehicle-mounted image, and adjusting the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image;

[0039] In this embodiment, a suitable ambient light source recognition model is selected, such as a convolutional neural network (CNN) based on deep learning or a traditional color space-based method. For applications with high real-time requirements, it is recommended to use the YOLO series model (such as YOLOv5) because it has a good balance between accuracy and speed. The distortion-optimized vehicle image is input into the selected ambient light source detection model. First, the image is adjusted to the input size required by the model (such as 640x640 pixels) and normalized to improve the recognition accuracy of the model. The model is run for reasoning to identify ambient light sources in the image, including natural light (such as sunlight) and artificial light (such as street lights, car lights, etc.). The model will output the category, position (bounding box coordinates) and confidence score of each light source. To improve the reliability of recognition, a confidence threshold (such as 0.5) is set to filter out recognition results with low confidence. The recognition results (light source type and bounding box information) are recorded in the database to ensure subsequent analysis and use. At the same time, the position and type of the identified light source are drawn on the image for visualization and analysis, helping to understand the impact of light source distribution on the image. Define spectral parameters, including color temperature, light intensity, etc., to describe the spectral characteristics of the ambient light source. Use the CIE standard light source model to calculate the corresponding spectral characteristics according to the identified light source type. The color temperature is calculated based on the average value of the RGB channels, using the standard color temperature calculation formula. Extract the characteristics of the identified light source, including its type, location, color temperature, and light intensity. These characteristics will be used for subsequent spectral parameter adjustment. Use the average brightness value under lighting conditions to calculate the light intensity to obtain more accurate parameters. Select appropriate spectral parameter adjustment algorithms, such as gamma correction, white balance adjustment, or color enhancement technology. These algorithms can dynamically adjust the spectral parameters of the image according to the characteristics of the ambient light source to achieve the best visual effect. According to the extracted ambient light source characteristics, dynamically adjust the spectral parameters of the distortion optimized vehicle image. Specific methods include: White balance adjustment: adjust the RGB channels of the image according to the color temperature of the light source to eliminate color cast. Light intensity adjustment: Based on the light source identification results, adjust the brightness of the image to ensure that the clarity and details of the image are preserved under different lighting conditions. Gamma correction: Adjust the contrast of the image according to the intensity of the ambient light source to maintain good visual effects under different lighting conditions.

[0040] Step S3: Calculate the histogram of each pixel of the dynamic spectrum adjustment image, and visualize the dynamic exposure range distribution to construct a dynamic exposure range distribution perception map;

[0041] In this embodiment, a suitable histogram calculation method is selected, which usually includes a grayscale histogram and an RGB histogram. For color images, it is recommended to calculate the histogram of each color channel separately so as to more comprehensively analyze the brightness and color distribution of the image. Traverse each pixel in the dynamic spectrum adjustment image and extract its RGB value (or grayscale value). By counting each pixel, its corresponding value is added to the corresponding bucket of the histogram. Usually, 256 buckets (0-255) are used to count the frequency of occurrence of each pixel value. After collecting the values ​​of all pixels, a histogram is constructed and normalized to convert the frequency into a relative frequency. This process ensures that the sum of the histogram is 1, which is convenient for subsequent analysis and comparison. The calculated histogram data is stored in a database, and a histogram image is generated for visualization. The shape of the histogram (such as a single peak, a double peak, etc.) is analyzed to determine the exposure and brightness distribution of the image. According to the histogram data, a dynamic exposure range is defined. Set the exposure range to the minimum and maximum values ​​of the pixel values ​​in the histogram. Usually, the quantiles of 0.1 and 0.9 are used to avoid the influence of extreme values. Calculate the overall exposure range of the dynamic spectrum adjustment image, including overexposed, underexposed, and normally exposed areas. The specific method is to identify the pixel value range of each area according to the distribution of the histogram. For example, set a threshold to consider pixel values ​​less than a certain value (such as 50) as underexposed and greater than a certain value (such as 205) as overexposed. Choose a suitable visualization method. Usually, heat maps or area maps are used to represent different types of exposure areas by color depth and area size. Heat maps can clearly show the distribution of exposure ranges for easy observation. Generate a dynamic exposure range distribution perception map based on the calculated exposure range data. Map the information of different exposure areas to the map, and use different colors to identify overexposed, underexposed, and normally exposed areas. For example, overexposed areas are represented by red, underexposed areas are represented by blue, and normally exposed areas are represented by green. Save the generated dynamic exposure range distribution perception map and record the relevant data parameters for subsequent analysis and processing. At the same time, a visualization tool is used to display the exposure range distribution perception map to help analyze the performance of the image under different lighting conditions and ensure the effectiveness of subsequent processing steps.

[0042] Step S4: performing region-by-region adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment according to the dynamic exposure range distribution perception map, and extracting optimal exposure compensation parameters;

[0043] In this embodiment, a region division method is selected according to the dynamic exposure range distribution perception map. A clustering algorithm (such as K-means or DBSCAN) is used to partition the image to ensure that each region has a certain similarity in exposure characteristics. The number of regions is set according to the actual situation, for example, the image is divided into 9 regions for more detailed compensation. Feature extraction is performed on each divided region, and the average brightness, contrast and exposure distribution of each region are calculated. For example, the average value and standard deviation of the pixel value of each region are calculated to evaluate its overall exposure. A threshold is set to identify overexposed, underexposed and normally exposed regions. The region division and its characteristics are recorded in a database, and the division results are displayed through a visualization tool to help analyze the exposure characteristics of each region. This step helps the subsequent adaptive exposure compensation calculation and ensures the accuracy of the region division. A suitable compensation parameter calculation model is selected, usually a linear model or a nonlinear model based on image features. According to the exposure characteristics of each region, the required exposure compensation parameters are calculated. For example, the compensation parameter range is set to [-1, 1], where a negative value indicates a reduced exposure and a positive value indicates an increased exposure. For each region, the exposure compensation parameters are calculated according to its exposure characteristics. For overexposed areas, the compensation parameters are set to negative values ​​to reduce the brightness; for underexposed areas, the compensation parameters are set to positive values ​​to increase the brightness. The calculated exposure compensation parameters for each area are recorded in the database, and statistical analysis is performed to calculate the average compensation value and standard deviation of each area. These results will provide an important basis for subsequent multi-parameter exposure compensation adjustment. Select a suitable multi-parameter exposure compensation adjustment algorithm, such as polynomial adjustment or adjustment method based on curve fitting. These algorithms can adjust the overall exposure of the image based on the calculated compensation parameters. For each area, the calculated exposure compensation parameters are applied to adjust the image. The linear transformation formula is used to adjust the brightness of the image, and the adjusted image is evaluated for quality. The image quality before and after adjustment is compared using indicators such as PSNR and SSIM. At the same time, the adjustment effects of different areas are analyzed to ensure that the exposure compensation of each area achieves the expected effect. Select a suitable optimal parameter extraction method, usually using a greedy algorithm or a weighted average method, which comprehensively considers the adjustment effect of each area and the overall image quality. Extract the optimal exposure compensation parameters based on the adjusted image quality evaluation results. This process is achieved by weighted averaging the compensation parameters of each area, with the weights set based on the area or importance of each area. The extracted optimal exposure compensation parameters are recorded in the database to ensure easy access for subsequent applications. At the same time, a visual chart is generated to show the relationship between the optimal compensation parameters and the adjustment effect of each area, so as to facilitate analysis and optimization.

[0044] Step S5: accurately locate the bounding boxes of each object one by one according to the real-time vehicle-mounted recorded images, and perform dynamic movement trajectory fitting to generate the movement trajectory of each dynamic object;

[0045] In this embodiment, an efficient object detection model is selected, such as YOLOv4 or Faster R-CNN, which has an excellent balance between speed and accuracy. According to the characteristics of the real-time vehicle-mounted recorded image, the hyperparameters of the model are adjusted to adapt to the dynamic environment when the vehicle is driving. Before the real-time vehicle-mounted image is input into the detection model, it needs to be preprocessed, including adjusting the image size (such as 640x640 pixels) and normalization (scaling the pixel value to the [0, 1] interval). In addition, data enhancement (such as random cropping, rotation and color transformation) is performed to improve the robustness of the model. Run the selected object detection model, infer the real-time image, identify all dynamic objects in the image, and generate a bounding box and corresponding category and confidence score for each object. Set a confidence threshold (such as 0.5) to only retain high-confidence detection results to reduce false detection. Perform non-maximum suppression (NMS) processing to remove bounding boxes with high overlap to ensure that each object has only one optimal bounding box. Set an overlap threshold (such as 0.3) and compare the intersection over union (IoU) of bounding boxes to determine the degree of overlap, so as to retain the most representative bounding boxes. Record the bounding box information (position, category, and confidence) of each object in the database, and draw the bounding box and label on the image for subsequent analysis and visualization. Display the detection results through visualization tools to help evaluate the accuracy and effectiveness of the detection. For each detected dynamic object, initialize its movement trajectory. Take the coordinates (x, y) of the center point of the bounding box of each object as the starting point and record the position change of each object in the time series. In the real-time image stream, use an object tracking algorithm (such as Kalman filter or CSRT) to track each object in consecutive frames. By analyzing the bounding box positions of the previous and next frames, update the center coordinates of each object to ensure that the position change of the object can be reflected in real time. According to the motion characteristics of the object, select a suitable motion model for trajectory fitting. For example, for a vehicle driving smoothly, a linear model is used; for a motion trajectory with large changes, a polynomial model or spline fitting method can be used to better capture the motion trajectory of the object. According to the collected object position data (the coordinates of the center point of the bounding box), the selected motion model is used to fit the trajectory. Through optimization techniques such as the least squares method, the motion trajectory of each dynamic object is fitted and the mathematical expression of the trajectory is generated. The movement trajectory data of each object (coordinate points, speed, etc.) is recorded in the database and a trajectory visualization diagram is generated. The motion trajectory of each object is displayed through visualization tools to help analyze its motion pattern and behavior, ensuring a deeper understanding of dynamic objects.

[0046] Step S6: Perform dynamic tracking exposure decision according to the moving trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

[0047] In this embodiment, based on the movement trajectory data of each dynamic object, the motion characteristics of the object, including speed, acceleration and path curvature, are analyzed. These motion characteristics will help determine the position and state of the object in the future frame, thereby affecting the exposure decision. For example, a threshold is set to identify dynamic changes when the speed changes by more than 10%, so as to make timely exposure adjustments. According to the motion trajectory of the object, its exposure requirements are evaluated. Fast-moving objects require higher exposure to ensure a bright picture, while slower-moving objects perform well under low exposure. Using the measured light intensity data and image histogram information, combined with the object speed, the baseline value of the exposure requirement is set. For example, when the object speed exceeds 5 m / s, the value of the exposure compensation parameter is increased to avoid the image being too dark. Design a suitable decision model, using a rule-based decision system or a machine learning model (such as a decision tree). This model will input the motion characteristics of the object (such as speed, direction) and the current ambient lighting conditions (such as light source type, light intensity), and output corresponding exposure adjustment suggestions. Set the output range of the model, such as the exposure compensation parameter between [-1, 1], to ensure the controllability of the decision. Run the decision model to make real-time exposure decisions for each dynamic object. Dynamically adjust the exposure settings based on the analysis results and model output. Record the exposure decision for each object and update the exposure parameters in real time in the image processing chain to ensure the stability of image quality. Build an end-to-end exposure control model that integrates the input data into the object's movement trajectory, current exposure settings, and optimal exposure compensation parameters. The model needs to have real-time response capabilities to make adjustments in a rapidly changing environment. Select a suitable control algorithm, such as a PID controller or a fuzzy controller. These algorithms can adjust the exposure settings in real time based on the deviation between the current exposure value and the target exposure value. Set the parameters of the PID controller (proportional, integral, and derivative coefficients), such as P=0.1, I=0.01, D=0.05, to ensure the stability and fast response of the control process. In the exposure control model, monitor the exposure level of the current image and compare it with the target exposure level. According to the control algorithm, the exposure compensation parameters that need to be adjusted are calculated in real time and applied to the image processing chain. For example, when it is detected that the current exposure value is lower than the target exposure value, increase the exposure compensation parameters to increase the image brightness. Monitor the quality of the adjusted image and evaluate the exposure effect of the image in real time. Use indicators such as PSNR and SSIM to evaluate whether the adjusted image quality meets expectations. At the same time, record the parameters and results of each exposure adjustment for subsequent optimization and model improvement. Based on the feedback results, continuously optimize the exposure control model and decision algorithm. Adjust the control parameters and model structure to improve the system's adaptability to dynamic environments and the accuracy of exposure control. Conduct regular experimental evaluations to ensure the stability and effectiveness of the model under different lighting and motion conditions.

[0048] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0049] Step S11: acquiring real-time vehicle recorded images based on the vehicle image sensor;

[0050] Step S12: Decomposing the real-time vehicle recorded image into time series frames, and extracting the vehicle image time series frames;

[0051] Step S13: Calculating the delay between adjacent frames of the vehicle-mounted image time sequence frames to generate a plurality of delay parameters between adjacent frames;

[0052] Step S14: performing inter-frame delay variation identification based on a plurality of adjacent inter-frame delay parameters to generate an image inter-frame delay variation feature;

[0053] Step S15: performing dynamic delay interpolation optimization on the real-time vehicle recorded image based on the delay variation characteristics between image frames to obtain an image with optimized delay between frames;

[0054] Step S16: Perform motion blur compensation on the inter-frame delay optimized image to construct a distortion optimized vehicle image.

[0055] In this embodiment, a high-resolution on-board image sensor is selected, and the resolution should generally reach 1080p (1920x1080) or higher to ensure the clarity of the recorded image. The sensor should have good dynamic range and low-light performance to adapt to different environmental conditions (such as night or cloudy days). Integrate the image sensor with the on-board computing unit to ensure that the data can be transmitted in real time. Use a USB interface or a MIPI interface to connect the sensor and the computing unit to ensure that the bandwidth for data transmission is sufficient, and a frame rate of at least 30fps is usually required to ensure smooth recording. Configure the software to achieve real-time image recording. By setting a suitable encoding format (such as H.264 or H.265), balance image quality and storage space to ensure that key data is not lost during the recording process. The real-time recorded image should be saved to a high-speed storage device, such as an SSD, to increase the write speed and avoid frame loss due to delays. While recording the image, record relevant metadata, including timestamps, GPS locations, vehicle speeds, etc., for subsequent analysis and processing. This information is obtained through on-board sensors or external GPS devices and stored together with the image data. Read data from the saved real-time vehicle recorded images and use video processing libraries (such as OpenCV) to decode the images. Split the entire video stream into individual frames for subsequent processing. Decompose the time series frames according to the set frame rate (for example, extract 30 frames per second). Save each extracted frame as an image file (such as JPEG or PNG format) and record its corresponding timestamp and other metadata. Preprocess each extracted time series frame, including denoising, adjusting brightness and contrast, etc., to improve the accuracy of subsequent analysis. Use Gaussian filtering or median filtering for denoising to ensure that the quality of the frame meets the analysis standard. Classify and mark the frames according to the timestamp and content of the extracted frames to ensure that subsequent processing steps can quickly locate specific frames. This process helps to quickly filter out images of a specific time period when needed. Choose an appropriate delay calculation method, which is usually based on the difference in timestamps. Determine the delay between each frame by calculating the difference in timestamps (in seconds) between adjacent frames. Traverse all extracted time series frames and generate delay parameters for adjacent frames based on the timestamps. Delay = time (n+1) − timestamp (n). Record the delay value of each pair of adjacent frames and store it in a data structure (such as a list or array). Perform statistical analysis on the calculated delay parameters and calculate statistical values ​​such as average delay, maximum delay, and minimum delay. These statistical values ​​can help understand the stability and consistency of image recording. Use graphical tools (such as Matplotlib) to visualize the delay parameters to show the changing trend of inter-frame delay. This visualization can help identify potential problems, such as sudden increases or large fluctuations in delay, indicating bottlenecks in data transmission or processing. Select an appropriate delay change identification algorithm, such as a threshold-based change detection method or a statistical-based change detection method.The threshold method sets a standard value, and when the delay exceeds this value, it is considered a change; the statistical method uses standard deviation or moving average method to identify abnormal points. By calculating the delay change between adjacent frames (for example, the difference between the delay of the current frame and the delay of the previous frame), a change feature is generated. This calculation can help identify the mutation point of the delay for subsequent processing. The identified delay changes are extracted, such as the change amplitude, change frequency, etc., and these features are combined with the corresponding timestamp and frame information and recorded in a data structure. The generated delay change features are analyzed to identify the patterns and trends of delay changes. These results are displayed through visualization tools to help understand the factors affecting delay changes. Select a suitable dynamic interpolation algorithm, such as linear interpolation, spline interpolation, or more complex deep learning-based interpolation methods. The selected algorithm should be able to effectively generate missing frames based on the delay change features. Set interpolation parameters based on the delay change features, such as selecting the interpolation window size and interpolation method. The size of the interpolation window is adjusted based on the frequency and amplitude of the delay change. For each pair of adjacent frames, the interpolation value is calculated based on the delay change features to generate an optimized image frame. The specific process includes using an interpolation algorithm to calculate the pixel values ​​of the missing frames and fill them into the image sequence. The quality of the interpolated optimized images is evaluated, including comparing the quality differences between the original images and the optimized images. Evaluation is performed using indicators such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Select a suitable motion blur compensation model, such as a frequency-domain based Wiener filter or a spatial-domain deconvolution method. These methods can effectively handle the blur caused by motion. By analyzing the motion trajectory in the inter-frame delay optimized image, the parameters of motion blur, such as the direction and degree of blur, are estimated. These parameters are obtained by calculating motion vectors or optical flow fields. The optimized image is motion-blur compensated using the estimated blur parameters. The selected compensation algorithm is applied to each image frame to reduce the blur effect and restore image details. The quality of the compensated image is evaluated to ensure that the motion blur compensation is effective. A combination of subjective and objective evaluation methods is used to ensure that the final distortion-optimized vehicle image has good clarity and detail.

[0056] In this embodiment, the specific steps of step S15 are:

[0057] Perform dynamic object visual recognition on inter-frame delay optimized images and mark multiple dynamic objects;

[0058] Perform dynamic optical flow estimation on multiple dynamic objects and extract motion vectors of dynamic objects;

[0059] Perform motion blur analysis on the inter-frame delay optimized image pair to generate motion blur data in the image;

[0060] According to the delay variation characteristics between image frames, the motion vector of the dynamic object is delayed and calculated to obtain the compensation parameters of the object motion vector with delay between frames.

[0061] Motion blur removal is performed on motion blur data in the image based on object motion vector compensation parameters of inter-frame delay, thereby generating a motion blur removal image;

[0062] Perform lens distortion detection on motion blur-removed images and extract sensor lens distortion data;

[0063] Perform geometric distortion optimization based on sensor lens distortion data to construct distortion-optimized vehicle images.

[0064] In this embodiment, a suitable computer vision model is selected for dynamic object recognition. Commonly used models include YOLO (You Only Look Once) and SSD (Single Shot Detector). These models can detect and mark dynamic objects in images in real time. According to actual needs, select the version of the model, such as YOLOv4 or YOLOv5, to ensure the accuracy and speed of detection. Input the inter-frame delay optimized image into the selected dynamic object detection model. Make sure that the size of the image meets the input requirements of the model. Usually, the image needs to be scaled to a specified size (such as 416x416 pixels). Before input, perform data enhancement (such as random cropping, rotation, etc.) to improve the robustness of the model. Run the detection model, infer the input image, and identify the dynamic objects therein. The results of the model output include the category, location (bounding box coordinates) and confidence score of the object. According to the confidence threshold (for example, 0.5), filter out the detection results with high confidence, and draw the bounding box and label on the image. Visualize the detection results, display the marked images, and save the marked information (such as object category, location, etc.) to a file or database for subsequent analysis and processing. Select a suitable optical flow estimation algorithm, such as the Lucas-Kanade method or the Horn-Schunck method. These algorithms can effectively calculate the motion vector of objects in the image and are suitable for dynamic object motion estimation in inter-frame delay optimized images. Extract adjacent frames from the inter-frame delay optimized images for optical flow calculation. Based on the detection results of dynamic objects, select the frames containing the target objects for processing to ensure the accuracy of optical flow estimation. Apply the optical flow algorithm to the selected adjacent frames to calculate the motion vector of each pixel. The output results are the horizontal and vertical motion components of each pixel, which can reflect the motion of the dynamic object in the image. Extract the motion vector of the dynamic object from the optical flow field, usually by calculating the average optical flow vector in the object bounding box area. Record the motion direction and speed of each dynamic object to provide a basis for subsequent analysis. Select a suitable motion blur detection algorithm, such as the Laplace operator method or frequency-domain based blur detection. These methods can effectively identify the motion blur area in the image and quantify the degree of blur. Use the selected algorithm to process the inter-frame delay optimized image to identify the motion blurred area. Determine the boundary and range of the blurred area by calculating the gradient or frequency characteristics of each pixel. Quantify the degree of blur in the blurred area, usually expressed as a blur intensity value. This is achieved by calculating the average gradient value or the energy of the frequency domain characteristics in the blurred area. The lower the value, the higher the degree of blur. Record the motion blur data in a database, saving the location information, blur intensity value, etc. of each blurred area. At the same time, display the blur analysis results through visualization tools for reference in subsequent processing. Extract the inter-frame delay change characteristics from the previous steps, including the delay parameters between adjacent frames. These features help understand the delay of motion vectors.Select a suitable compensation algorithm, usually based on linear interpolation or polynomial interpolation, and use the delay characteristics to correct the motion vector. The compensation algorithm should be able to adjust the direction and magnitude of the motion vector according to the delay parameters. Apply the delay compensation algorithm to the motion vector of each dynamic object and calculate the compensated motion vector. In the specific calculation process, the delay value is combined with the motion vector to generate a new motion vector. Record the compensated motion vector to ensure that the motion information of each dynamic object is updated. These compensated parameters will be used for subsequent motion blur removal and geometric distortion optimization. Select a suitable motion blur removal algorithm, such as deconvolution method or algorithm based on blind deconvolution. These algorithms can effectively restore image details lost due to motion blur. Use the compensated motion vector as an input parameter to set the relevant parameters of the removal algorithm. Ensure that the blur removal algorithm can accept the motion information of dynamic objects so as to better restore the image. Apply the removal algorithm to the motion blur data and perform deconvolution. According to the direction and magnitude of the motion vector, the algorithm will repair the blurred area and generate a clear motion blur-removed image. Evaluate the quality of the removed image and quantify the restoration effect using indicators such as PSNR and SSIM. Adjust the parameters of the elimination algorithm based on the evaluation results to improve the quality of the final image. Select a suitable lens distortion detection algorithm. Common methods include checkerboard calibration or feature point-based detection methods. These methods effectively identify geometric distortion in images. Use a standard checkerboard image for calibration, extract feature points in the image, and calculate the difference between the actual coordinates and the ideal coordinates. By comparison, identify the radial and tangential distortion parameters of the lens. Build a lens distortion model based on the extracted distortion data. A polynomial model is usually used to describe the degree of distortion, including the calculation and adjustment of distortion coefficients. Record the extracted lens distortion data in a database to ensure that it is used in subsequent processing. At the same time, visualize the distortion detection results to help understand the impact of distortion. Select a suitable geometric distortion optimization algorithm, such as parallax correction or remapping technology. These technologies can effectively correct geometric distortion in images and restore the real scene. Use the extracted lens distortion data as input parameters to set the relevant parameters of the geometric optimization algorithm. Ensure that the algorithm can use the distortion model to correct the image. Apply the geometric optimization algorithm to the motion blur-free image and perform image remapping. According to the distortion parameters, the position of each pixel is adjusted to generate an image with optimized distortion. The quality of the optimized image is evaluated to ensure that the geometric distortion is effectively corrected. The evaluation indicators include clarity, detail retention, etc. Finally, the optimized vehicle image is saved for subsequent use and analysis.

[0065] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0066] Step S21: performing ambient light source recognition on the distortion-optimized vehicle-mounted image and marking the real-time ambient light source;

[0067] Step S22: Calculating the color temperature of the ambient light source according to the real-time ambient light source to generate a color temperature feature of the ambient light source;

[0068] Step S23: performing real-time ambient light intensity distribution calculation on the distortion-optimized vehicle-mounted image to generate a temporal distribution feature of the ambient light intensity;

[0069] Step S24: performing real-time scene spectrum feature evolution according to the ambient light source color temperature feature and the ambient light intensity temporal distribution feature to generate real-time scene spectrum feature;

[0070] Step S25: adjusting the dynamic image spectrum parameters of the distortion-optimized vehicle-mounted image based on the real-time scene spectrum characteristics to generate a dynamic spectrum adjusted image.

[0071] In this embodiment, a suitable ambient light source detection algorithm is selected. Common methods include color histogram-based detection and deep learning models (such as YOLO or Faster R-CNN). For real-time ambient light source recognition, deep learning models are more effective due to their high accuracy and adaptability. The distortion-optimized vehicle-mounted image is used as input data to ensure that its format meets the requirements of the detection model. It is usually necessary to scale the image to a specified size (such as 416x416 pixels) and perform standardization to improve the recognition accuracy of the model. Run the selected ambient light source detection model to infer the input image. The model will output the identified light source position (bounding box) and type (such as natural light, street lights, car lights, etc.). According to the confidence threshold (for example, 0.5), filter out high-confidence results and draw bounding boxes and labels on the image. Record the recognition results (light source type and position) in the database to ensure subsequent processing. At the same time, the marked image is displayed through a visualization tool to help analyze and understand the distribution of light sources. Use a standard color temperature calculation formula, such as a color temperature estimation formula based on RGB channels. Color temperature = R / G × 1000. Select a suitable model or algorithm for more complex color temperature estimation, such as McCamy's approximation formula. Extract RGB color values ​​from the identified ambient light sources. For each light source, extract the average RGB value of the corresponding area based on its bounding box to obtain accurate color information. Use the selected color temperature calculation method to substitute the extracted RGB values ​​into the formula to calculate the color temperature of each ambient light source. Record the calculation results, including the color temperature value of each light source and its corresponding timestamp. Save the calculated color temperature characteristics of the ambient light source to the database for subsequent analysis. At the same time, analyze the color temperature changes of different light sources to help understand the impact of ambient light on the image. Select a suitable ambient light intensity calculation method, such as grayscale-based light intensity calculation, or use light intensity data collected by the sensor. Usually, light intensity is calculated using the following formula: Light intensity = 0.299 × R + 0.587 × G + 0.114 × B. Perform grayscale conversion on the distortion-optimized vehicle image to calculate the light intensity value of each pixel. Use the above formula to directly calculate the light intensity from the RGB value and generate a light intensity distribution map. Combined with time series data, analyze the light intensity distribution at different time points. Record the light intensity change characteristics in each time period, including the maximum, minimum and average values ​​of the light intensity, to generate time series distribution characteristics. Record the generated ambient light intensity time series distribution characteristics in the database, and display the light intensity change curve through visualization tools to help understand the changes in the light environment in different time periods. Select a suitable spectral feature model. Common methods include spectral models based on color temperature and light intensity. Use the CIE standard light source model to calculate the spectral characteristics of the scene in combination with color temperature and light intensity. Based on the previously calculated ambient light source color temperature characteristics and light intensity time series distribution characteristics, use the selected spectral model to generate the spectral characteristics of the scene. The specific calculation method includes mapping the color temperature to the spectral curve and calculating the light intensity values ​​at different wavelengths.Combined with real-time data, analyze the changes in the scene's spectral characteristics. Record the spectral characteristics at different time points, including spectral distribution, peak wavelength, and light intensity changes, to help understand the evolution of spectral characteristics. Record the generated real-time scene spectral characteristics in the database, and display the spectral change curve through visualization tools to help understand the impact of ambient light on the scene. Select a suitable dynamic spectral adjustment algorithm. Common methods include adjustment algorithms based on color correction and adjustment methods based on image fusion. Consider using common technologies such as gamma correction and white balance adjustment. Set the relevant parameters of the adjustment algorithm based on the generated real-time scene spectral characteristics. Including color adjustment ratio, contrast, and brightness. These parameters should be fine-tuned according to the actual environmental spectral characteristics. Apply the selected dynamic spectral adjustment algorithm to the distortion-optimized vehicle-mounted image to adjust the spectral parameters. According to the real-time spectral characteristics, dynamically adjust the color and brightness of the image to ensure that the image is more natural and realistic. Evaluate the quality of the adjusted image, and use visual evaluation and objective indicators (such as PSNR and SSIM) to compare the difference between the original image and the adjusted image. Adjust the parameters of the algorithm based on the evaluation results to improve the quality of the final image.

[0072] In this embodiment, refer to Figure 4 , is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0073] Step S31: Calculate the histogram of each pixel of the dynamic spectrum adjustment image, and extract the histogram feature of each pixel;

[0074] Step S32: dividing the dynamic spectrum adjustment image into multiple regions to generate a multi-region image;

[0075] Step S33: performing histogram morphology recognition according to the histogram features of each pixel point and performing exposure degree analysis to generate exposure classification data;

[0076] Step S34: performing regional exposure identification on the multi-region image according to the exposure classification data to generate exposure characteristics of each region, wherein the exposure characteristics of each region include an overexposed region, an underexposed region, and an acceptable dynamic range region;

[0077] Step S35: Visualize the dynamic exposure range distribution of the exposure characteristics of each area and construct a dynamic exposure range distribution perception map.

[0078] In this embodiment, a suitable histogram calculation method is selected, and a grayscale histogram or an RGB histogram is usually used. For color images, the histogram of each color channel is calculated separately to facilitate the analysis of the color distribution of each pixel. Traverse each pixel in the dynamic spectral adjustment image and extract its RGB value (or grayscale value). Based on these values, the histogram data structure is updated, and 256 buckets (0-255) are usually used to count the frequency of occurrence of each pixel value. For each pixel, its position in the histogram and the corresponding frequency are recorded. These features will provide important information about the brightness and color distribution of the image, especially when analyzing exposure. The histogram features of each pixel are stored in a data structure, such as a two-dimensional array or a database. Ensure the accessibility of the data to facilitate the analysis and processing of subsequent steps. Select a suitable region division strategy, commonly used methods include grid division method or region division method based on image segmentation (such as K-means clustering or superpixel segmentation). The selected strategy should be determined according to the image content and analysis requirements. The dynamic spectral adjustment image is processed. If the grid division method is used, the image is divided into several rectangular areas; if the image segmentation method is used, the algorithm is used to identify areas with similar characteristics. Make sure that the size and shape of each area are suitable for subsequent analysis. Extract the features of each divided area, including average color, brightness, and histogram. These features will help in subsequent exposure analysis. Record the features of each area in a database to ensure easy access in subsequent steps. At the same time, mark the area for visualization and analysis. Select a suitable histogram morphology recognition algorithm, commonly used are threshold method, morphological operation, etc. The exposure level is determined by analyzing the shape of the histogram (such as single peak, double peak, etc.). Calculate the exposure level of each pixel based on its histogram features. Use the distribution of light intensity (such as overexposure above 255 and underexposure below 0) for classification. Set an appropriate threshold, usually referring to empirical values, such as 0.1 and 0.9 quantiles. Based on the calculation results, classify each pixel as overexposed, underexposed, or normally exposed. Record these classification results and mark them in the attributes of each pixel for subsequent analysis. Store the exposure classification data in a database to ensure data accessibility. Display the histogram and its morphology through visualization tools to facilitate intuitive analysis of exposure conditions. Select an appropriate regional exposure identification method, such as a statistical analysis-based method or a regional feature aggregation method. Consider determining the exposure of a region by calculating the exposure classification ratio of pixels in the region. Perform statistics on the pixel exposure classification data of each divided region, and calculate the number and ratio of overexposed, underexposed, and normally exposed pixels. Generate exposure feature data for each region, including the proportion of each type of region. Based on the calculation results, generate exposure features for each region, including areas marked as overexposed, underexposed, and acceptable dynamic range areas. Record these features for subsequent analysis and processing. Store the exposure features of each region in a database to ensure access in subsequent steps.At the same time, use visualization tools to display regional exposure characteristics so that the exposure conditions of different regions can be intuitively understood. Choose appropriate visualization methods, such as heat maps or regional maps, which can effectively display the distribution of regional exposure characteristics. Heat maps use color depth to represent the proportion of different exposure characteristics. Process the exposure characteristic data of each region and prepare the required format for visualization. Ensure that the data can reflect the exposure conditions of each region, especially overexposure and underexposure. Use data visualization tools (such as Matplotlib, Tableau, etc.) to generate dynamic exposure range distribution perception maps. Use different colors to represent different types of regions according to the exposure characteristics of the region to ensure that the graphics are clear and easy to understand. Analyze the generated dynamic exposure range distribution perception maps to evaluate their effectiveness in practical applications. Adjust visualization parameters based on feedback to improve the readability and accuracy of the graphics.

[0079] In this embodiment, step S4 includes the following steps:

[0080] Step S41: performing adaptive exposure compensation calculation for each region according to the dynamic exposure range distribution perception map to generate an exposure compensation parameter range for each region;

[0081] Step S42: performing multiple compensation parameter quantitative processing according to the exposure compensation parameter range of each area to generate multiple exposure compensation parameters;

[0082] Step S43: performing multi-parameter exposure compensation adjustment on the dynamic spectrum adjustment image according to multiple exposure compensation parameters to obtain multiple exposure compensated images;

[0083] Step S44: comprehensively evaluating the compensation effects of the multiple exposure compensation images to generate evaluation values ​​of the multiple compensation images; and extracting optimal exposure compensation parameters based on the evaluation values ​​of the multiple compensation images.

[0084] In this embodiment, the dynamic exposure range distribution perception map is analyzed to identify the exposure characteristics of each area. These areas are classified into overexposure, underexposure or acceptable dynamic range areas according to exposure. The exposure of each area is recorded to ensure that there is enough data to support subsequent compensation calculations. A suitable adaptive exposure compensation algorithm is selected, such as an algorithm based on local contrast enhancement or a linear transformation method. These algorithms can dynamically adjust the exposure compensation parameters according to regional characteristics to achieve the best visual effect. For each area, the range of exposure compensation parameters is calculated according to its exposure characteristics. For overexposed areas, a negative compensation value is set to reduce the brightness; for underexposed areas, a positive compensation value is set to increase the brightness. A standardized exposure adjustment range (such as -1 to +1) is used to make fine adjustments according to the specific conditions of regional characteristics. The exposure compensation parameter range of each area is recorded in a database, and a visualization chart of the compensation parameter range is generated to help analyze the exposure compensation requirements of different areas. This visualization uses a heat map to show the compensation degree of different areas. Select a suitable compensation parameter generation method, such as uniform sampling, random sampling, or distribution-based sampling methods. These methods can effectively generate multiple compensation parameters for subsequent exposure compensation adjustments. For the exposure compensation parameter range of each region, the range is subdivided into multiple discrete compensation values. For example, if the compensation range of a region is [-0.5, +0.5], it is divided into 10 equally spaced values ​​(such as -0.5, -0.4, -0.3, ..., 0.4, 0.5). Multiple exposure compensation parameters are generated according to the subdivided compensation range. These parameters will be used for subsequent multi-parameter exposure compensation adjustment to ensure that different exposure conditions can be covered during the adjustment process. The generated multiple exposure compensation parameters are stored in a database to ensure that these parameters can be easily accessed in subsequent steps. In addition, the region information corresponding to each parameter is recorded for easy analysis and evaluation. Select a suitable exposure compensation adjustment algorithm, and commonly used methods include linear adjustment, gamma correction, or color balance adjustment. These algorithms can dynamically adjust the brightness and contrast of the image according to the set compensation parameters. For each generated exposure compensation parameter, the selected adjustment algorithm is applied one by one. Ensure that the details and quality of the image can be maintained during the adjustment process. For example, for each compensation parameter, the brightness of the image is adjusted using a linear transformation formula. After applying the compensation parameters, the corresponding multiple exposure compensated images are generated. These images will show different exposure compensation effects, providing a rich data basis for subsequent comprehensive evaluation. Save the generated multiple exposure compensation images to the database or file system, and record the compensation parameters corresponding to each image for subsequent evaluation and analysis. Select appropriate evaluation indicators to conduct a comprehensive evaluation of the compensated images. Commonly used indicators include peak signal-to-noise ratio (PSNR), structural similarity index (SSIM) and visual quality score (VQS). These indicators can effectively reflect the clarity and detail retention of the image. For each exposure compensation image generated, calculate its evaluation index value.Each image is compared with the original image to obtain its PSNR, SSIM, VQS and other index values, and generate evaluation results. Based on the calculated evaluation values, a comprehensive evaluation is performed on multiple exposure compensated images. The exposure compensation parameters corresponding to the image with the highest evaluation value are selected as the optimal compensation parameters. This process is achieved through simple sorting or weighted averaging. The evaluation results and optimal exposure compensation parameters are recorded in the database to ensure easy access later. At the same time, the evaluation results are displayed through visualization tools to help analyze the effects of different compensation parameters and the optimal parameters finally selected.

[0085] In this embodiment, step S5 includes the following steps:

[0086] Step S51: classifying multiple dynamic objects into different types to generate dynamic object types;

[0087] Step S52: accurately locate the bounding boxes of each object one by one according to the type of the dynamic object, and generate multiple dynamic object bounding boxes;

[0088] Step S53: performing continuous frame tracking on multiple dynamic object bounding boxes to generate a real-time position of each dynamic bounding box;

[0089] Step S54: performing dynamic movement trajectory fitting according to the real-time position of each dynamic boundary box to generate the movement trajectory of each dynamic object.

[0090] In this embodiment, a suitable object classification model is selected, and commonly used ones include YOLO (You Only Look Once), SSD (Single Shot Detector) or a method based on a convolutional neural network (CNN). According to the specific application scenario, a high-precision and good real-time model, such as YOLOv5, is selected to ensure that the type of dynamic objects can be quickly identified. An image containing multiple dynamic objects (such as an image after exposure compensation) is input into the selected classification model. According to the model requirements, the image size is adjusted to a specified size (such as 640x640 pixels), and preprocessing (such as normalization, data enhancement, etc.) is performed to improve the classification accuracy. The object classification model is run to infer the input image. The model will output the type (such as vehicle, pedestrian, animal, etc.), location information (bounding box coordinates) and confidence score of each identified object. According to the set confidence threshold (such as 0.5), the classification results with high confidence are screened out, and the bounding box and type label are drawn on the image. The classification results (object type and corresponding bounding box information) are recorded in the database for subsequent analysis and processing. At the same time, the marked images are displayed through visualization tools to help analyze the distribution of dynamic objects. Select a suitable bounding box adjustment method, such as non-maximum suppression (NMS), to eliminate overlapping bounding boxes and ensure that each object has only one optimal bounding box. This method can effectively improve the accuracy of the bounding box. Extract the bounding box coordinates of each object (such as the coordinates of the upper left corner and the lower right corner) from the object classification results. Use the NMS algorithm to process the bounding boxes with high overlap, set a suitable threshold (such as 0.3), and remove redundant boxes. Accurately locate the bounding box of each dynamic object to ensure that the bounding box can tightly surround the object. Use image segmentation technology (such as Mask R-CNN) to further optimize the shape and position of the bounding box to make it more consistent with the actual contour of the object. Store the bounding box information of multiple dynamic objects generated in the database, and record the object type of each bounding box. Display the optimized bounding box through a visualization tool to ensure that the position of each object in the image is clearly visible. Select an appropriate object tracking algorithm, such as Kalman filter, CSRT (Discriminative Correlation Filter with Channel and Spatial Reliability), or Deep SORT (Simple Online and Realtime Tracking). These algorithms can stably track the position of the object in consecutive frames. Initialize the tracker in the first frame and input the bounding box information and object type. This process ensures that the tracker can identify and track each dynamic object. Traverse subsequent frames and update the bounding box position of each dynamic object using the selected tracking algorithm. The tracker will dynamically adjust the position of the bounding box based on the object's motion characteristics (such as speed, acceleration).Record the real-time position of each dynamic object in the database for subsequent analysis. At the same time, display the tracking results through visualization tools to ensure that the movement trajectory of the dynamic object in the image is clearly visible. Select a suitable trajectory fitting method, such as polynomial fitting, spline fitting, or Bezier curve fitting. These methods can generate smooth trajectories based on real-time position data to reflect the movement path of the object. Collect the real-time position data of each dynamic object, including the center point coordinates of the bounding box in each frame. Ensure the integrity and timing of the data to facilitate subsequent trajectory fitting. Apply the selected trajectory fitting method for calculation based on the collected real-time position data. Generate the motion trajectory of each object and record the fitting parameters and the mathematical expression of the trajectory. Record the generated dynamic object movement trajectory in the database and display the trajectory diagram through visualization tools. Ensure that the motion path of each object is clearly visible to help analyze the object's motion behavior and pattern.

[0091] In this embodiment, step S6 includes the following steps:

[0092] Step S61: Calculate the moving speed of each dynamic object moving trajectory to generate the moving speed of the dynamic object;

[0093] Step S62: adjusting the local exposure frequency according to the moving speed of the dynamic object to generate a local exposure adjustment frequency;

[0094] Step S63: analyzing the dynamic object illumination change rate for each dynamic object moving trajectory to generate the dynamic object illumination change rate;

[0095] Step S64: fine-tuning the local exposure amplitude based on the dynamic object illumination change rate to generate a local exposure fine-tuning amplitude;

[0096] Step S65: making dynamic tracking exposure decisions based on the illumination change rate of the dynamic object and the local exposure fine-tuning amplitude, and constructing a local dynamic tracking exposure strategy;

[0097] Step S66: performing end-to-end exposure control according to the local dynamic tracking exposure control strategy and the optimal exposure compensation parameters to construct a dynamic exposure control model.

[0098] In this embodiment, a suitable moving speed calculation method is selected, and the Euclidean distance formula is usually used to calculate the displacement of the object between consecutive frames. The time interval is set (such as a frame rate of 30fps and a time interval of 1 / 30 second) to ensure the accuracy of the speed calculation. The real-time position data of each dynamic object in the consecutive frames is collected, including the coordinates of the center point of the bounding box (x, y). These coordinates are recorded for subsequent calculations to ensure the timing and integrity of the data. For each dynamic object, its speed between two frames is calculated. The calculated moving speed of the dynamic object is recorded in the database, and statistical analysis is performed to calculate the average speed, maximum speed and minimum speed of each object. These results will provide an important basis for subsequent exposure adjustment. The local exposure frequency adjustment strategy is determined according to the moving speed of the dynamic object. Generally, faster moving objects require more frequent exposure adjustments to ensure image quality. A threshold is set according to the object speed, for example, the speed is divided into three levels: slow (<2 m / s), medium (2-5 m / s) and fast (>5 m / s). Different speed levels correspond to different exposure frequency adjustment strategies. According to the moving speed of the object and its classification, the corresponding local exposure adjustment frequency is generated. For example, slower objects are set to adjust once per frame, while faster objects are set to adjust once every 0.5 frames. The calculated local exposure adjustment frequency is recorded in the database, and the exposure adjustment frequency of objects at different speeds is displayed through visualization tools for easy understanding and analysis. Define the rate of illumination change as the amplitude of the change in illumination intensity per unit time. Select an appropriate illumination change analysis method, such as time series analysis of light intensity. In each frame, record the illumination intensity around the dynamic object, including the intensity values ​​of natural light and ambient light. Use sensor data or extract light intensity directly from the image. For each dynamic object, analyze its illumination intensity changes in consecutive frames, illumination change rate = ,in, and are the light intensity in two consecutive frames, is the time interval. The calculated illumination change rate is recorded in the database and statistically analyzed to calculate the average illumination change rate of each object. These results will provide an important basis for subsequent exposure fine-tuning. Set the local exposure amplitude fine-tuning strategy according to the illumination change rate of the dynamic object. Generally, objects with higher illumination change rates require more significant exposure adjustments to avoid image distortion. Set a threshold according to the illumination change rate, for example, classify the rate into three levels: low (<0.5), medium (0.5-1.5) and high (>1.5). Different illumination change rates correspond to different exposure fine-tuning amplitudes. Generate the corresponding exposure fine-tuning amplitude according to the illumination change rate of the object and its classification. For example, the fine-tuning amplitude of an object with a lower illumination change rate is set to ±0.1, while that of an object with a higher rate is set to ±0.5. Record the calculated local exposure fine-tuning amplitude in the database to ensure easy access in subsequent steps. At the same time, the exposure fine-tuning amplitude of objects with different illumination change rates is displayed through visualization tools for easy analysis. Select a suitable dynamic tracking exposure decision algorithm. Common methods include rule-based decision systems or machine learning methods (such as decision trees). The selected algorithm should be able to respond quickly to real-time data. Collect the illumination change rate and local exposure fine-tuning amplitude data of dynamic objects as input parameters. Ensure the timing and integrity of the data to facilitate subsequent decision analysis. Apply the selected decision algorithm to make dynamic tracking exposure decisions based on the collected data. Adjust the exposure settings in real time based on the illumination change rate and exposure fine-tuning amplitude to ensure the best image quality. Record the dynamic tracking exposure decision results in the database to ensure the availability of subsequent analysis. At the same time, analyze the effects of different decision strategies to optimize the subsequent exposure adjustment strategy. Design a suitable dynamic exposure control model, which should combine the local dynamic tracking exposure control strategy with the optimal exposure compensation parameters. The model should be able to automatically adjust the exposure under real-time conditions. Integrate the local dynamic tracking exposure control strategy and the optimal exposure compensation parameters as input to ensure that the model can obtain the latest exposure information in real time. Select a control algorithm (such as PID controller or fuzzy control) for real-time exposure adjustment. Based on the input parameters, the model calculates and outputs adjustment instructions in real time to ensure the stability of image quality. Evaluate the constructed dynamic exposure control model and compare the image quality before and after adjustment using subjective evaluation and objective indicators (such as PSNR, SSIM). The model parameters are adjusted based on the evaluation results to improve the quality and stability of the final image.

[0099] In this embodiment, a wide dynamic range adjustment system based on a vehicle-mounted image sensor is provided, which is used to execute the wide dynamic range adjustment method based on a vehicle-mounted image sensor as described above, including:

[0100] A distortion optimization module is used to obtain real-time vehicle-mounted recorded images based on the vehicle-mounted image sensor; dynamic delay interpolation optimization and motion blur compensation are performed on the real-time vehicle-mounted recorded images to construct distortion-optimized vehicle-mounted images;

[0101] The spectrum adjustment module is used to identify the ambient light source of the distortion-optimized vehicle-mounted image, and adjust the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image;

[0102] The exposure distribution perception module is used to calculate the histogram of each pixel of the dynamic spectrum adjustment image, visualize the dynamic exposure range distribution, and construct a dynamic exposure range distribution perception map;

[0103] An exposure compensation module is used to perform adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment for each area according to the dynamic exposure range distribution perception map, and extract the optimal exposure compensation parameters;

[0104] The object tracking module is used to accurately locate the bounding box of each object based on the real-time vehicle-mounted recorded images, and to fit the dynamic movement trajectory to generate the movement trajectory of each dynamic object;

[0105] The dynamic tracking exposure module is used to make dynamic tracking exposure decisions according to the movement trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

[0106] The present invention uses the on-board image sensor in scenes of high-speed driving or fast movement, and the image will be distorted due to delay. Through delayed interpolation optimization, the time dislocation and inter-frame delay of the image are reduced, the image is smooth and consistent with the timing, and the image distortion caused by time deviation is reduced. Motion blur is a common problem in high-speed or dynamic scenes. Through motion estimation and blur compensation algorithms, the blurred part is effectively restored to make the image details clearer. Motion blur compensation technology is particularly important to improve the perception of the real-time environment by the automatic driving system and ensure that key objects and road signs can remain clear even in motion. Distortion optimization and motion blur compensation work together to ensure the quality of the on-board image, and can still provide clear and accurate image data for subsequent processing even in fast driving or complex scenes. By real-time identification of the light source in the image, the system can distinguish different lighting conditions (such as daylight, cloudy days, nighttime, artificial light sources, etc.). The identification of ambient light sources helps the system understand the lighting environment of the current scene, and further adjusts the color and brightness of the image to avoid overexposure or color distortion. By adjusting the image's spectral parameters such as color temperature, saturation, and contrast, the system optimizes the image's visual effects to ensure that image details are clearly presented under different lighting environments. In particular, in strong lighting changes (such as day-to-night transitions, light reflections, etc.), it can effectively avoid overexposure or underexposure and maintain the image's natural color and layering. This dynamic spectral adjustment ensures that the image can present the best color and brightness in different environments, allowing the vehicle perception system to obtain accurate information in complex lighting environments (such as tunnels, night driving, and reflected light). The brightness distribution in the image is accurately captured by calculating the brightness value histogram of each pixel. This enables the system to identify overexposed and underexposed areas and make targeted adjustments to these areas. A dynamic exposure range distribution perception map is constructed to intuitively display the exposure status of each area in the image, helping the system to fully understand the brightness distribution of the image and discover potential exposure problems. This visualization method enables the system to more accurately determine which areas require exposure compensation during real-time driving and formulate corresponding adjustment strategies to ensure that details can be clearly presented in scenes with a wide dynamic range. The system performs adaptive exposure compensation for different areas (such as highlight areas, shadow areas, etc.) based on the exposure distribution perception map. By analyzing the exposure requirements of each area, the system intelligently adjusts parameters such as exposure time and ISO gain to ensure balanced exposure of the entire image. Not only does it adjust the exposure time, but it also combines multiple parameters such as ISO gain to perform exposure compensation, so that the brightness and details of each area in the image are optimally balanced. By dynamically compensating for each area, the phenomenon of some areas in the image being too dark or too bright is avoided, ensuring that details in high dynamic range scenes are retained and improving the visual effect of the overall image.Through high-precision object detection algorithms (such as YOLO, Faster R-CNN, etc.), the system can accurately detect and mark various objects in the image (such as pedestrians, vehicles, traffic signs, etc.). The accurate positioning of the object provides the basis for subsequent object tracking and exposure adjustment. Combined with the detection results of the object, the tracking algorithm (such as Kalman filtering, SORT, etc.) is used to fit the motion trajectory of the object in real time. This motion trajectory tracking can help the system understand the movement trend of the object and provide more accurate information for dynamic exposure control. Object tracking technology enables the system to understand the state of the dynamic object in front in real time when the vehicle is driving at high speed and respond in time, thereby improving the safety and accuracy of the autonomous driving system. Based on the motion trajectory of the object, the system can quickly adjust the exposure of fast-moving objects (such as a moving car), while using a smooth exposure strategy for slow-moving objects (such as pedestrians). This dynamic adjustment ensures that the objects in the image are always clearly visible in high-speed driving environments. By combining the dynamic trajectory of the object and the optimal exposure compensation parameters, the system controls the exposure of the entire image end-to-end to ensure that each dynamic object can adapt to different lighting conditions during movement. For example, a fast-moving car will maintain clear details under strong backlight. This precise dynamic exposure control is critical for autonomous driving systems, ensuring that the system always provides clear and accurate visual information in complex and changing lighting environments, thereby improving driving safety.

[0107] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0108] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A wide dynamic range adjustment method based on a vehicle-mounted image sensor, characterized in that: The following steps are involved: Step S1: acquiring a real-time vehicle-mounted recorded image based on a vehicle-mounted image sensor; performing dynamic delay interpolation optimization and motion blur compensation on the real-time vehicle-mounted recorded image to construct a distortion-optimized vehicle-mounted image; Step S2: identifying the ambient light source for the distortion-optimized vehicle-mounted image, and adjusting the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image; Step S3: Calculate the histogram of each pixel of the dynamic spectrum adjustment image, and visualize the dynamic exposure range distribution to construct a dynamic exposure range distribution perception map; Step S4: performing region-by-region adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment according to the dynamic exposure range distribution perception map, and extracting optimal exposure compensation parameters; Step S5: accurately locate the bounding boxes of each object one by one according to the real-time vehicle-mounted recorded images, and perform dynamic movement trajectory fitting to generate the movement trajectory of each dynamic object; Step S6: Perform dynamic tracking exposure decision according to the moving trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

2. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: acquiring real-time vehicle recorded images based on the vehicle image sensor; Step S12: Decomposing the real-time vehicle recorded image into time series frames, and extracting the vehicle image time series frames; Step S13: Calculating the delay between adjacent frames of the vehicle-mounted image time sequence frames to generate a plurality of delay parameters between adjacent frames; Step S14: performing inter-frame delay variation identification based on a plurality of adjacent inter-frame delay parameters to generate an image inter-frame delay variation feature; Step S15: performing dynamic delay interpolation optimization on the real-time vehicle recorded image based on the delay variation characteristics between image frames to obtain an image with optimized delay between frames; Step S16: Perform motion blur compensation on the inter-frame delay optimized image to construct a distortion optimized vehicle image.

3. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 2, characterized in that: The specific steps of step S15 are: Perform dynamic object visual recognition on inter-frame delay optimized images and mark multiple dynamic objects; Perform dynamic optical flow estimation on multiple dynamic objects and extract motion vectors of dynamic objects; Perform motion blur analysis on the inter-frame delay optimized image pair to generate motion blur data in the image; According to the delay variation characteristics between image frames, the motion vector of the dynamic object is delayed and calculated to obtain the compensation parameters of the object motion vector with delay between frames. Motion blur removal is performed on motion blur data in the image based on object motion vector compensation parameters of inter-frame delay, thereby generating a motion blur removal image; Perform lens distortion detection on motion blur-removed images and extract sensor lens distortion data; Perform geometric distortion optimization based on sensor lens distortion data to construct distortion-optimized vehicle images.

4. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing ambient light source recognition on the distortion-optimized vehicle-mounted image and marking the real-time ambient light source; Step S22: Calculating the color temperature of the ambient light source according to the real-time ambient light source to generate a color temperature feature of the ambient light source; Step S23: performing real-time ambient light intensity distribution calculation on the distortion-optimized vehicle-mounted image to generate a temporal distribution feature of the ambient light intensity; Step S24: performing real-time scene spectrum feature evolution according to the ambient light source color temperature feature and the ambient light intensity temporal distribution feature to generate real-time scene spectrum feature; Step S25: adjusting the dynamic image spectrum parameters of the distortion-optimized vehicle-mounted image based on the real-time scene spectrum characteristics to generate a dynamic spectrum adjusted image.

5. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: Calculate the histogram of each pixel of the dynamic spectrum adjustment image and extract the histogram feature of each pixel; Step S32: dividing the dynamic spectrum adjustment image into multiple regions to generate a multi-region image; Step S33: performing histogram morphology recognition according to the histogram features of each pixel point and performing exposure degree analysis to generate exposure classification data; Step S34: performing regional exposure identification on the multi-region image according to the exposure classification data to generate exposure characteristics of each region, wherein the exposure characteristics of each region include an overexposed region, an underexposed region, and an acceptable dynamic range region; Step S35: Visualize the dynamic exposure range distribution of the exposure characteristics of each area and construct a dynamic exposure range distribution perception map.

6. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: performing adaptive exposure compensation calculation for each region according to the dynamic exposure range distribution perception map to generate an exposure compensation parameter range for each region; Step S42: performing multiple compensation parameter quantitative processing according to the exposure compensation parameter range of each area to generate multiple exposure compensation parameters; Step S43: performing multi-parameter exposure compensation adjustment on the dynamic spectrum adjustment image according to multiple exposure compensation parameters to obtain multiple exposure compensated images; Step S44: comprehensively evaluating the compensation effects of the multiple exposure compensation images to generate evaluation values ​​of the multiple compensation images; and extracting optimal exposure compensation parameters based on the evaluation values ​​of the multiple compensation images.

7. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: classifying multiple dynamic objects into object types to generate dynamic object types; Step S52: accurately locate the bounding boxes of each object one by one according to the type of the dynamic object, and generate multiple dynamic object bounding boxes; Step S53: performing continuous frame tracking on multiple dynamic object bounding boxes to generate a real-time position of each dynamic bounding box; Step S54: performing dynamic movement trajectory fitting according to the real-time position of each dynamic bounding box to generate a movement trajectory of each dynamic object.

8. The wide dynamic range adjustment method based on the vehicle-mounted image sensor according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: Calculate the moving speed of each dynamic object moving trajectory to generate the moving speed of the dynamic object; Step S62: adjusting the local exposure frequency according to the moving speed of the dynamic object to generate a local exposure adjustment frequency; Step S63: analyzing the dynamic object illumination change rate for each dynamic object moving trajectory to generate the dynamic object illumination change rate; Step S64: fine-tuning the local exposure amplitude based on the dynamic object illumination change rate to generate a local exposure fine-tuning amplitude; Step S65: making dynamic tracking exposure decisions based on the illumination change rate of the dynamic object and the local exposure fine-tuning amplitude, and constructing a local dynamic tracking exposure strategy; Step S66: performing end-to-end exposure control according to the local dynamic tracking exposure control strategy and the optimal exposure compensation parameters to construct a dynamic exposure control model.

9. A wide dynamic range adjustment system based on a vehicle-mounted image sensor, characterized in that: The method for performing the wide dynamic range adjustment method based on the vehicle-mounted image sensor as claimed in claim 1 comprises: A distortion optimization module is used to obtain real-time vehicle-mounted recorded images based on the vehicle-mounted image sensor; dynamic delay interpolation optimization and motion blur compensation are performed on the real-time vehicle-mounted recorded images to construct distortion-optimized vehicle-mounted images; The spectrum adjustment module is used to identify the ambient light source of the distortion-optimized vehicle-mounted image, and adjust the dynamic image spectrum parameters to generate a dynamic spectrum-adjusted image; The exposure distribution perception module is used to calculate the histogram of each pixel of the dynamic spectrum adjustment image, visualize the dynamic exposure range distribution, and construct a dynamic exposure range distribution perception map; An exposure compensation module is used to perform adaptive exposure compensation calculation and multi-parameter exposure compensation adjustment for each area according to the dynamic exposure range distribution perception map, and extract the optimal exposure compensation parameters; The object tracking module is used to accurately locate the bounding box of each object based on the real-time vehicle-mounted recorded images, and to fit the dynamic movement trajectory to generate the movement trajectory of each dynamic object; The dynamic tracking exposure module is used to make dynamic tracking exposure decisions according to the movement trajectory of each dynamic object, and perform end-to-end exposure control based on the optimal exposure compensation parameters to build a dynamic exposure control model.

Images