Camera adaptive adjustment method and system, and storage medium

Abstract

The invention relates to the technical field of camera adjustment, in particular to a camera adaptive adjustment method and system and a storage medium. The method comprises the following steps: acquiring a multi-source environment sensing data set and an initial environment image; performing environment feature recognition on the initial environment image according to the multi-source environment sensing data set to obtain a scene environment feature map; performing semantic scene segmentation on the scene environment feature map to obtain a semantic scene macroscopic feature map; performing time sequence attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map; performing feature constraint fusion on the semantic scene macroscopic feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map; and performing parameter adjustment on the camera according to the target vehicle-scene motion semantic map to obtain a camera parameter adjustment strategy set. According to the invention, the performance and the image quality of the camera in a changeable environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of camera adjustment, and in particular to a camera adaptive adjustment method, system and storage medium. Background Art

[0002] Traditional cameras perform significantly differently under different lighting conditions. In strong or low light environments, the quality of images captured by the camera will drop significantly, resulting in loss of details or overexposure, which is a serious problem for application scenarios that require clear images for accurate analysis. For example, in traffic monitoring, low light conditions at night or in tunnels can seriously affect the accuracy of vehicle and pedestrian recognition, increasing safety risks. Secondly, when faced with dynamic scenes, existing cameras are often unable to adjust parameters in real time to adapt to rapidly changing environments. For example, in urban traffic monitoring, the rapid movement of vehicles and changing traffic flow require the camera to be able to quickly adjust parameters such as focus and exposure to maintain image clarity and real-time performance. However, most camera systems lack this dynamic adjustment capability, resulting in the inability to provide stable and reliable monitoring images in complex dynamic scenes. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a camera adaptive adjustment method, system and storage medium to solve at least one of the above technical problems.

[0004] To achieve the above object, a camera adaptive adjustment method comprises the following steps:

[0005] Step S1: Acquire a multi-source environmental sensing data set and an initial environmental image; perform environmental feature recognition on the initial environmental image according to the multi-source environmental sensing data set to obtain a scene environmental feature map;

[0006] Step S2: performing semantic scene segmentation on the scene environment feature map to obtain a semantic scene macro feature map; performing temporal attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map; performing feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map;

[0007] Step S3: adjusting the camera parameters according to the target vehicle-scene motion semantic graph to obtain a camera parameter adjustment strategy set; screening the camera parameter adjustment strategy set to obtain an initial parameter adjustment scheme; predicting and compensating the initial parameter adjustment scheme to obtain a real-time parameter adjustment strategy;

[0008] Step S4: Based on the real-time parameter adjustment strategy, multi-parameter linkage modeling is performed on the camera to obtain a camera parameter combination model; the camera parameter combination model is evaluated by a fuzzy controller to obtain parameter linkage evaluation data; feedback control adjustment is performed according to the parameter linkage evaluation data to obtain an optimized parameter control scheme;

[0009] Step S5: intelligently reconstruct the camera parameters and collect images according to the optimized parameter control scheme to obtain a real-time monitoring image; perform image quality assessment on the real-time monitoring image to obtain image quality assessment data; update the optimized parameter control scheme according to the image quality assessment data to obtain the final camera configuration parameters.

[0010] The present invention can accurately capture the environmental features of the scene and generate a scene environmental feature map by acquiring a multi-source environmental sensing data set and an initial environmental image and performing environmental feature recognition. This enables the camera to adapt to different lighting conditions, whether it is a strong light or weak light environment, to maintain image quality, reduce detail loss and overexposure, thereby improving the performance of the camera under various lighting conditions and the availability of the image. By performing semantic segmentation and temporal attention target tracking on the scene environmental feature map, a target vehicle-scene motion semantic map can be generated, which not only improves the understanding of static environmental elements, but also enhances the recognition and tracking capabilities of dynamic targets such as vehicles and pedestrians. This is particularly important for application scenarios such as traffic monitoring that require accurate recognition and tracking of targets, and can significantly improve the accuracy and response speed of the monitoring system. By generating a camera parameter adjustment strategy set, screening and predictive compensation thereof, a real-time parameter adjustment strategy is obtained. This means that the camera can intelligently adjust its parameters, such as focal length and exposure, according to real-time environmental changes and target dynamics to maintain image clarity and real-time performance. This intelligent adjustment capability enables the camera to better adapt to rapidly changing dynamic scenes and provide stable and reliable monitoring images. Through multi-parameter linkage modeling and fuzzy controller evaluation, parameter linkage evaluation data can be obtained, and feedback control adjustment can be performed based on these data to obtain an optimized parameter control scheme. This not only improves the efficiency and accuracy of camera parameter adjustment, but also ensures that the camera can work in the best state under various complex environments, further improving the quality of monitoring images. The present invention also includes image quality evaluation of real-time monitoring images, and updating the parameters of the parameter control scheme based on the evaluation data to obtain the final camera configuration parameters. This enables the camera system to self-optimize, continuously adapt to environmental changes, and ensure long-term and stable provision of high-quality monitoring images.

[0011] Preferably, the present invention further provides a camera adaptive adjustment system, which is used to execute the camera adaptive adjustment method as described above, and the camera adaptive adjustment system comprises:

[0012] The environmental feature recognition module is used to obtain a multi-source environmental sensor data set and an initial environmental image; perform environmental feature recognition on the initial environmental image according to the multi-source environmental sensor data set to obtain a scene environmental feature map;

[0013] The target tracking module is used to perform semantic scene segmentation on the scene environment feature map to obtain a semantic scene macro feature map; perform temporal attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map; perform feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map;

[0014] The adjustment strategy generation module is used to adjust the camera parameters according to the target vehicle-scene motion semantic graph to obtain a camera parameter adjustment strategy set; screen the camera parameter adjustment strategy set to obtain an initial parameter adjustment plan; predict and compensate the initial parameter adjustment plan to obtain a real-time parameter adjustment strategy;

[0015] The feedback control adjustment module is used to perform multi-parameter linkage modeling on the camera based on the real-time parameter adjustment strategy to obtain the camera parameter combination model; perform fuzzy controller evaluation on the camera parameter combination model to obtain parameter linkage evaluation data; perform feedback control adjustment based on the parameter linkage evaluation data to obtain an optimized parameter control scheme;

[0016] The parameter updating module is used to perform intelligent parameter reconstruction and image acquisition of the camera according to the optimized parameter control scheme to obtain real-time monitoring images; perform image quality assessment on the real-time monitoring images to obtain image quality assessment data; and update the parameters of the optimized parameter control scheme according to the image quality assessment data to obtain the final camera configuration parameters.

[0017] The present invention realizes comprehensive and dynamic adjustment of camera parameters by integrating environmental feature recognition module, target tracking module, adjustment strategy generation module, feedback control adjustment module and parameter update module. The system can quickly identify environmental features and track targets, generate detailed semantic maps and motion feature maps, thereby improving the resolution of monitoring images and the accuracy of target recognition. By adjusting camera parameters in real time, the system can adapt to various lighting and dynamic environments to ensure that high-quality images can be captured under different conditions. The system generates real-time parameter adjustment strategies through screening and predictive compensation, effectively allocates camera resources, and improves the operating efficiency of the monitoring system. Through multi-parameter linkage modeling and fuzzy controller evaluation, the system can achieve fine control of parameters and enhance the stability and reliability of the camera in complex environments. The system can continuously update the parameter control scheme based on image quality evaluation data, so that the camera system has the ability to self-optimize and maintain the best monitoring effect for a long time.

[0018] Preferably, a computer-readable storage medium stores a camera adaptive adjustment method as described above which can be loaded and executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Other features, objects and advantages of the present invention will become more apparent from reading the detailed description with reference to the following drawings:

[0020] Figure 1 A schematic flow chart of the steps of a camera adaptive adjustment method according to an embodiment is shown.

[0021] Figure 2 A detailed flow chart of step S3 of an embodiment is shown.

[0022] Figure 3 A detailed flowchart diagram of step S37 of an embodiment is shown. DETAILED DESCRIPTION

[0023] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0025] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0026] To achieve this, please refer to Figures 1 to 3 The present invention provides a camera adaptive adjustment method, comprising the following steps:

[0027] Step S1: Acquire a multi-source environmental sensing data set and an initial environmental image; perform environmental feature recognition on the initial environmental image according to the multi-source environmental sensing data set to obtain a scene environmental feature map;

[0028] Step S2: performing semantic scene segmentation on the scene environment feature map to obtain a semantic scene macro feature map; performing temporal attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map; performing feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map;

[0029] Step S3: adjusting the camera parameters according to the target vehicle-scene motion semantic graph to obtain a camera parameter adjustment strategy set; screening the camera parameter adjustment strategy set to obtain an initial parameter adjustment scheme; predicting and compensating the initial parameter adjustment scheme to obtain a real-time parameter adjustment strategy;

[0030] Step S4: Based on the real-time parameter adjustment strategy, multi-parameter linkage modeling is performed on the camera to obtain a camera parameter combination model; the camera parameter combination model is evaluated by a fuzzy controller to obtain parameter linkage evaluation data; feedback control adjustment is performed according to the parameter linkage evaluation data to obtain an optimized parameter control scheme;

[0031] Step S5: intelligently reconstruct the camera parameters and collect images according to the optimized parameter control scheme to obtain a real-time monitoring image; perform image quality assessment on the real-time monitoring image to obtain image quality assessment data; update the optimized parameter control scheme according to the image quality assessment data to obtain the final camera configuration parameters.

[0032] In this embodiment, using environmental sensors and camera hardware, first obtain a multi-source environmental sensing data set, including a light intensity data stream and a depth measurement data stream, and an initial environmental image. These data are transmitted to a data processing server in real time through the API of the sensor and camera. Using the deep learning framework TensorFlow, the initial environmental image and the multi-source environmental sensing data set are analyzed to identify the scene environment feature map. This involves training a convolutional neural network (CNN) to process image data and extract environmental features. Using the image segmentation algorithm in the OpenCV library, the scene environment feature map is semantically segmented to obtain a semantic scene macro feature map. This includes applying a pre-trained deep learning model to identify different objects and scene parts in the image. A temporal attention network is implemented using the PyTorch framework to perform temporal attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map. This network can identify and track target vehicles in image sequences. The semantic scene macro feature map is feature-constrained fused with the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map. Based on the target vehicle-scene motion semantic map, a genetic algorithm is used to generate a camera parameter adjustment strategy set. This algorithm can simulate the natural selection process to optimize the camera parameter settings. From the parameter adjustment strategy set, the greedy selection strategy is used to select the initial parameter adjustment scheme, which is the best estimate based on the current environment and scene requirements. The initial parameter adjustment scheme is predicted and compensated, and the ARIMA model is used to predict environmental changes and adjust the parameters accordingly to obtain the real-time parameter adjustment strategy. Based on the real-time parameter adjustment strategy, MATLAB is used to perform multi-parameter linkage modeling to obtain the camera parameter combination model. This involves establishing mathematical relationships between parameters and simulating how they jointly affect camera performance. The camera parameter combination model is evaluated by fuzzy controller, and MATLAB's Fuzzy Logic Toolbox is used to deal with uncertainty and nonlinear problems to obtain parameter linkage evaluation data. Based on the parameter linkage evaluation data, a PID controller is used for feedback control adjustment to obtain an optimized parameter control scheme. This controller can adjust parameters based on the error signal to achieve the expected image quality. According to the optimized parameter control scheme, the camera is intelligently reconstructed through the camera API and real-time monitoring images are collected. Image quality evaluation is performed on the real-time monitoring images, and image quality evaluation data is obtained using image quality evaluation tools such as IMQE (Image Quality Expert). According to the image quality evaluation data, the genetic algorithm is used again to update the parameters of the optimization parameter control scheme to obtain the final camera configuration parameters, ensuring that the camera can provide the best image quality in various environments.

[0033] Preferably, step S1 comprises the following steps:

[0034] Step S11: collecting environmental parameters of the environment where the camera is located to obtain a multi-source environmental sensing data set; wherein the multi-source environmental sensing data set includes a light intensity data stream and a depth measurement data stream;

[0035] Specifically, a group of sensors, including light sensors and depth cameras, can be deployed in the environment where the camera is located. The light sensor is responsible for capturing the light intensity. It can monitor the changes in ambient light in real time and convert the light signal into an electrical signal. These electrical signals are then converted into a digital format through an analog-to-digital converter (ADC) to obtain a light intensity data stream. At the same time, the depth camera obtains the depth information of the environment by emitting infrared light and measuring the time it takes to be reflected back, generating a depth measurement data stream. These data streams contain the distance information from each pixel to the camera.

[0036] Step S12: collecting a range image of the environment where the camera is located to obtain an initial environment image;

[0037] Specifically, a digital camera can be used to collect a range image of the environment. This camera is equipped with a wide-angle lens that can cover a wider field of view, thereby capturing a more comprehensive image of the environment. The collected image is saved in a secure digital storage medium, such as a solid-state drive (SSD) or a high-speed SD card, and the initial environment image is finally obtained.

[0038] Step S13: performing compensation illumination enhancement on the initial environment image according to the illumination intensity data stream to obtain an environment enhanced illumination feature map;

[0039] Specifically, image processing software, such as Adobe Photoshop or GIMP, can be used to analyze the light intensity data stream. These data streams provide real-time information about the ambient light level, including the intensity and direction of the light. Import the initial ambient image into the image processing software, and use the software's histogram tool to analyze the image's brightness distribution to determine which areas in the image are too dark or too bright. Based on the results of the light intensity data stream and histogram analysis, manually or automatically adjust the image's brightness and contrast. For example, if the data shows that the ambient light is dim, increase the overall brightness of the image, or use the "Shadow / Highlight" tool to restore dark details while avoiding overexposing bright areas, and finally obtain an ambient enhanced lighting feature map.

[0040] Step S14: performing distance gradient calibration on the initial environment image according to the depth measurement data stream to obtain an environment calibration depth measurement map;

[0041] Specifically, please refer to the sub-steps of step S14 for the detailed implementation process of this embodiment.

[0042] Step S15: performing feature fusion on the environment enhanced illumination feature map and the environment calibration depth measurement map to obtain a scene environment feature map.

[0043] Specifically, image processing software, such as OpenCV, can be used to extract features from the environment enhancement illumination feature map and the environment calibration depth measurement map. For the environment enhancement illumination feature map, color and texture features are extracted; for the environment calibration depth measurement map, edge and contour features are extracted. Corresponding feature points are found in the two images through feature matching algorithms, such as SIFT (Scale Invariant Feature Transform) or ORB (Oriented FAST and Rotated BRIEF). Since the depth map and the illumination map may have slight differences in perspective or scale, the feature point matching results are used to align the two images to ensure that they are spatially consistent. Alignment can be achieved through affine transformation or perspective transformation, depending on the degree of difference between the images. Based on the alignment, image fusion techniques, such as Laplacian pyramid fusion or Poisson fusion, are used to fuse the two images into a scene environment feature map.

[0044] The present invention enhances the comprehensiveness of image acquisition by collecting light intensity and depth measurement data, significantly improves the image quality under different lighting conditions, reduces overexposure or underexposure, and improves the accuracy of depth information. The optimization of feature fusion technology further improves the scene understanding ability, and is more accurate in semantic segmentation and target tracking. The reliability of the monitoring system is improved by accurate environmental parameter acquisition and image calibration, especially reducing false alarms and missed alarms in key areas such as security monitoring.

[0045] Preferably, step S14 comprises the following steps:

[0046] Step S141: performing bilateral filtering on the initial environment image according to the depth measurement data stream to obtain a filtered environment image;

[0047] Specifically, the depth measurement data stream can be aligned with the initial environment image to ensure that the spatial resolution and coordinate system of the two are consistent. This involves scaling or cropping the depth image to match the size of the environment image. Set up the OpenCV environment and import the necessary libraries. Prepare the initial environment image and the corresponding depth measurement data stream, ensuring that both are loaded into memory. Adjust the parameters of the bilateral filter based on the depth information provided by the depth measurement data stream. The diameter parameter (D) is set to 9 to determine the size of the filter window. The standard deviation of the color space (sigmaColor) is set to 75 to control the weight of color similarity. The standard deviation of the coordinate space (sigmaSpace) is also set to 75 to control the weight of spatial proximity. These parameter settings take into account the impact of depth information on image noise. Use the OpenCV cv2.bilateralFilter function to perform a bilateral filtering operation with the initial environment image and the set parameters as input. The function will iterate over each pixel of the image and perform a weighted average of the surrounding pixels based on the depth measurement data stream and color information to generate a filtered environment image. After filtering, check the filtered environment image to ensure that the noise is effectively reduced while the edges and details of the image are preserved. Pay special attention to areas with large depth changes, such as object edges, to ensure that these areas remain clear after filtering. If the filtering result is not ideal, such as blurred edges or ineffective noise removal, adjust the parameters of the bilateral filter according to the characteristics of the depth measurement data stream and re-filter.

[0048] Step S142: Calculate the depth gradient of the filtered environment image to obtain a spatial depth gradient map;

[0049] Specifically, the Sobel operator can be used to calculate the depth gradient because it can effectively detect edge and texture information in the image. Use the cv2.Sobel function in the OpenCV library to calculate the gradient. Set the parameters of the function, including the input filtered environment image, the derivative order in the x and y directions (select the 1st order derivative), and the depth of the result image (select CV_32F, i.e. 32-bit floating point number). Calculate the gradient of the image in the x and y directions respectively. For each direction, the cv2.Sobel function will calculate the first-order derivative of the image and generate two gradient images, representing the gradient intensity in the x and y directions respectively. In order to obtain the gradient magnitude of each pixel, use the cv2.magnitude function, which takes the gradient images in two directions as input, outputs the gradient magnitude of each pixel, and generates a spatial depth gradient map.

[0050] Step S143: constructing a depth calibration transformation matrix according to the spatial depth gradient map to obtain an image depth calibration parameter set;

[0051] Specifically, the spatial depth gradient map can be loaded using image processing software (such as MATLAB or OpenCV) and its depth change characteristics can be analyzed. Depth discontinuities and edges in the image are identified. The stereo vision calibration method in the camera calibration technology is used, which involves using known reference objects (such as a chessboard) to determine the intrinsic and extrinsic parameters of the camera. Use a stereo matching algorithm (such as semi-global matching or block matching) to estimate the disparity map between the two views. The disparity map provides displacement information of the pixel points in the depth direction. Based on the disparity map and the known camera parameters, the image depth calibration parameter set is calculated. This includes the construction of a depth transformation matrix, which will be used to convert the disparity map into a depth map.

[0052] Step S144: performing perspective transformation correction on the filtered environment image using the image depth calibration parameter set to obtain an environment depth correction image;

[0053] Specifically, an image depth calibration parameter set and a filtered environment image can be loaded. These parameters include the depth transformation matrix and the intrinsic parameters of the camera. Using image processing software (such as OpenCV), call the perspective transformation function (such as cv2.perspectiveTransform) to apply the depth calibration parameter set to the initial environment image. During the perspective transformation process, a transformation matrix and a region of interest (ROI) of the image need to be provided. The transformation matrix is ​​provided by the depth calibration parameter set, and the ROI can be the entire image or a specific area in the image. The software performs perspective transformation correction on the filtered environment image based on the provided transformation matrix and ROI to obtain an environment depth correction map. This process involves recalculating the position of each pixel in the image to compensate for the distortion caused by the camera perspective. The corrected image (environment depth correction map) is compared with the original image to verify the effect of the correction. Check whether the geometric distortion of the image is effectively corrected and whether the depth information of the image is consistent with expectations.

[0054] Step S145: reconstructing the environment depth correction map in three dimensions to obtain an environment calibration depth measurement map.

[0055] Specifically, you can use professional 3D reconstruction software such as MeshLab or CloudCompare, which can process depth images and generate 3D point clouds or meshes. Import the environmental depth correction map into the 3D reconstruction software. This image contains depth information corrected by perspective transformation, providing accurate input data for 3D reconstruction. Set the necessary reconstruction parameters in the software. This includes the camera's intrinsic parameters (such as focal length, principal point coordinates) and extrinsic parameters (such as camera position and rotation), which are usually obtained through the camera calibration process. Use the software's 3D reconstruction function to convert the 2D depth correction map into a 3D point cloud. This process involves converting the depth value of each pixel to a point in 3D space and building a point cloud based on the camera's viewpoint. From the point cloud data, you can use the software's mesh generation function to create a 3D mesh or surface. This mesh represents the 3D structure of the environment and can be used for further analysis or visualization. Finally, verify the accuracy of the reconstruction by visualizing the 3D model. You can observe the model from different angles to check if there are obvious distortions or missing parts. If the model is as expected, then it is an environmental calibration depth measurement map.

[0056] The present invention processes the depth measurement data stream through bilateral filtering, effectively improving the accuracy and smoothness of the depth data, reducing noise and retaining edge information. The resolution of the depth information is enhanced by depth gradient calculation, and the scene depth changes can be understood in detail. The accuracy of the calibration process is optimized by constructing a depth calibration transformation matrix and an image depth calibration parameter set, and the reliability of the depth information is improved. The geometric accuracy of the image is improved and distortion is reduced through perspective transformation correction, while the authenticity of the depth measurement map is enhanced by three-dimensional space reconstruction.

[0057] Preferably, step S2 comprises the following steps:

[0058] Step S21: performing scene decomposition on the scene environment feature map to obtain an environment scene scale feature set;

[0059] Specifically, the image processing software OpenCV and the machine learning library scikit-learn can be used for scene decomposition. The scene environment feature map is segmented using the segmentation algorithm in OpenCV to identify different objects and backgrounds. , such as K-means clustering or watershed algorithm, to distinguish different areas in the image. These algorithms can segment an image into multiple parts based on features such as color and texture. For each segmented area, scale features are extracted, such as the area, perimeter, shape descriptors (such as circularity, rectangularity), etc. of the area. These features describe the scale and shape characteristics of each area. The extracted scale features are stored in an environmental scene scale feature set, with each area corresponding to a feature vector.

[0060] Step S22: Calculate the regional importance of the scene environment feature map based on the environment scene scale feature set to obtain an environment scene attention distribution map;

[0061] Specifically, OpenCV can be used to calculate the importance of regions. A series of criteria are defined to evaluate the importance of each region, such as the size of the region, the complexity of the shape, the degree of association with a specific target, etc. For each region in the feature set, its importance score is calculated according to the defined criteria. For example, regions with larger areas are given higher scores, and regions with complex shapes also receive higher scores. Based on the importance score of each region, an environmental scene attention distribution map is generated. In this distribution map, the color or brightness of each region represents its importance score, and the darker the color or the higher the brightness, the higher its importance. Finally, check the environmental scene attention distribution map to ensure that the distribution of regional importance in the map is as expected. Pay special attention to whether key areas such as pedestrians, vehicles, etc. are correctly highlighted in the map.

[0062] Step S23: performing sliding window feature extraction on the environmental scene attention distribution map to obtain a scene local feature data set;

[0063] Specifically, the Python programming language can be used in conjunction with the OpenCV library to implement sliding window feature extraction. Define a fixed-size window, such as 32x32 pixels, which will slide on the attention distribution map to cover the entire image. The size and step size of the window (the number of pixels the window moves each time) are determined according to the resolution of the image and the scale of the features you want to extract. As the window slides, a series of features are calculated for all pixels in the window, including color histograms, texture features (such as gray-level co-occurrence matrices), shape features, etc. These features can describe the local properties of the area within the window. For each window position, the extracted features are stored as a feature vector, and these feature vectors are combined into a scene local feature dataset. This scene local feature dataset contains the local features of all window positions in the image.

[0064] Step S24: performing scene region segmentation on the scene environment feature map according to the environment scene attention distribution map and the scene local feature data set to obtain a semantic scene macro feature map;

[0065] Specifically, Python and OpenCV libraries can be used in combination with image segmentation algorithms, such as graph-based segmentation or deep learning methods, to segment scene regions. First, the attention values ​​in the environmental scene attention distribution map are used as weights and combined with the feature vectors in the local feature dataset. This allows regions with higher attention to be prioritized during the segmentation process. The weighted local features and attention information are used to define the boundaries of the regions. For example, a graph-based segmentation algorithm can be used to represent the image as a graph, where nodes represent pixels and edges represent the similarity between pixels. Regions are merged or segmented based on the attention and feature similarity of the nodes. For each segmented region, semantic features are further extracted, such as the category label of the region (e.g., person, vehicle, building). This can be achieved by training a classifier (such as a support vector machine or a deep neural network) that classifies regions based on the local feature dataset. Finally, a semantic scene macro-feature map is constructed, which represents the semantic information and macro-features of each region in the scene. In this map, each region is labeled with its corresponding semantic label and macro-feature. The semantic scene macro-feature map is checked to ensure that the segmented regions accurately reflect the semantic structure of the scene and that the boundaries of the regions are clear and reasonable.

[0066] Step S25: generating a temporal feature sequence for the scene environment feature map to obtain a temporal feature set of a moving target;

[0067] Specifically, the Python programming language can be used with the OpenCV library and the NumPy library to process the scene environment feature map, and the SciPy library can be used to analyze the time series data. First, a series of temporally continuous scene environment feature maps are obtained. These images can be captured by the camera at fixed time intervals. For each frame of the image, the image processing algorithm is used to extract the features of the moving target, such as edges, corners, optical flow, etc. These features can describe the movement and change of the target over time. The features of each moving target are tracked and compared between consecutive frames to generate a time series feature sequence. For example, an optical flow algorithm is used to track the position change of the target in consecutive frames to obtain time series features such as speed and acceleration. The time series feature sequences of all moving targets are stored in a moving target time series feature set, and each target corresponds to a time series feature sequence. This dataset contains the time series information of all moving targets in the scene.

[0068] Step S26: Modeling the spatiotemporal attention mechanism for the moving target temporal feature set to obtain the target vehicle attention weight map;

[0069] Specifically, the spatiotemporal attention mechanism can be implemented using the Python programming language in conjunction with a deep learning framework such as TensorFlow or PyTorch. First, the temporal feature sequence is integrated with the spatial features (such as the shape, size, color, etc. of the target) to form a spatiotemporal feature set. A spatiotemporal attention network model is designed that can automatically learn which features are most important for the recognition and tracking of target vehicles in different scenes and at different time points. A convolutional neural network (CNN) is used to extract spatial features, and a recurrent neural network (RNN) or a long short-term memory network (LSTM) is used to process temporal features. By training the spatiotemporal attention network model, the attention weight of each pixel or region is obtained. These weights represent the degree of attention of the model to each position, and high-weighted regions indicate that the model considers these regions to be more important for the recognition of the target vehicle. The attention weights are applied to the scene environment feature map to generate a target vehicle attention weight map. In this map, regions of different colors or brightness represent different levels of attention weights.

[0070] Step S27: locating the moving target time series feature set based on the target vehicle attention weight map to obtain a target vehicle motion feature map;

[0071] Specifically, the Python programming language can be used with the deep learning framework TensorFlow and the computer vision library OpenCV to process the target vehicle attention weight map and the moving target temporal feature set. The target vehicle attention weight map is merged with the moving target temporal feature set to determine the position of the target vehicle in consecutive frames. The attention weight map provides the importance information of each pixel, which is used as a weighting factor to strengthen the positioning of the target vehicle in the temporal feature set. The tracking algorithms in OpenCV, such as Kalman filtering or Mean-Shift, are combined with the attention weights to track the target vehicle. These algorithms can locate the target in each frame and predict its position in the next frame. For each frame, motion features such as speed, direction, and acceleration are extracted based on the position and temporal features of the target vehicle. These features are used to generate a target vehicle motion feature map, which displays the motion information of the target vehicle in space.

[0072] Step S28: performing feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map.

[0073] Specifically, feature fusion can be performed using the Python programming language in conjunction with the image processing library OpenCV and the machine learning library scikit-learn. First, the semantic scene macro feature map and the target vehicle motion feature map are preprocessed, including normalization and feature alignment, to ensure that the two feature maps are consistent in spatial resolution and feature scale. The two feature maps are combined using feature fusion techniques, such as weighted fusion, graph-based fusion, or deep learning fusion networks. For example, a deep learning network can be designed with two feature maps as input and a fused target vehicle-scene motion semantic map as output. During the fusion process, feature constraints such as spatial consistency, semantic consistency, and motion consistency are applied to ensure that the fused feature map is logically reasonable. For example, ensure that the motion features correspond to the semantic features in spatial position. The fused feature map contains the macro semantic information of the scene and the motion information of the target vehicle, generating a target vehicle-scene motion semantic map. This map provides a comprehensive perspective for understanding the dynamic relationship between the scene and the target vehicle.

[0074] The present invention enhances the scene understanding ability by deeply analyzing the scene environment feature map, and can accurately identify key areas and understand the scene structure. The efficiency and accuracy of feature extraction are improved by sliding window feature extraction. By combining the attention distribution map and the local feature data set, the scene area segmentation is optimized and the semantic scene macro feature map is generated. The accuracy of moving target tracking is improved by the generation of temporal feature sequences and the modeling of spatiotemporal attention mechanisms, especially in complex dynamic scenes. Target positioning based on the attention weight map further improves the accuracy of the target vehicle motion feature map. The deep fusion of the semantic scene macro feature map and the target vehicle motion feature map is achieved through the feature constraint fusion technology, and the target vehicle-scene motion semantic map is generated, which enhances the comprehensive analysis capability of static and dynamic targets.

[0075] Preferably, step S3 comprises the following steps:

[0076] Step S31: performing spatiotemporal feature encoding on the target vehicle-scene motion semantic graph to obtain a scene state vector set;

[0077] Specifically, Python programming language can be used with deep learning library PyTorch to process spatiotemporal feature encoding. PyTorch provides flexible tensor operations and automatic differentiation functions, which are suitable for complex spatiotemporal feature encoding. Extract key spatiotemporal features from the target vehicle-scene motion semantic graph, including the position, velocity, acceleration of the target vehicle, and static and dynamic features in the scene. Design a spatiotemporal encoding network that can map the extracted spatiotemporal features into a high-dimensional space to form a scene state vector. This network includes convolutional layers to extract spatial features and recurrent layers (such as LSTM or GRU) to extract temporal features. For each time point, a scene state vector is obtained through the spatiotemporal encoding network, and these vectors are aggregated to form a scene state vector set. This set can represent the state of the scene at different time points. Check the scene state vector set to ensure that each vector can accurately reflect the scene state at the corresponding time point and that the vector set can capture the dynamic changes of the scene.

[0078] Step S32: constructing a dual network architecture for the scene state vector set to obtain a parameter adjustment strategy-evaluation network; wherein the parameter adjustment strategy-evaluation network includes a parameter action generation network and a state value evaluation network;

[0079] Specifically, the dual network architecture can be constructed using Python programming language and deep learning library PyTorch. Two deep neural networks are designed: a parameter action generation network and a state value evaluation network. The parameter action generation network is responsible for generating camera parameter adjustment actions based on the scene state vector, while the state value evaluation network is responsible for evaluating the value of these actions. This network includes multiple fully connected layers, with the scene state vector as input and the action space of parameter adjustment as output. Supervised learning is used to train this network so that it can generate effective parameter adjustment actions. This network also includes multiple fully connected layers, with the scene state vector and parameter adjustment actions as input and the value scores of these actions as output. Reinforcement learning is used to train this network so that it can evaluate the long-term value of actions. The two networks are integrated into a dual network architecture, in which the output of the parameter action generation network is used as one of the inputs of the state value evaluation network. In this way, the state value evaluation network can not only evaluate the value of the action, but also provide feedback to guide the generation process of the parameter action generation network. The performance of the dual network architecture is verified by a simulation environment to ensure that the parameter adjustment strategy-evaluation network can generate effective parameter adjustment strategies and can be dynamically adjusted according to changes in scene states.

[0080] Step S33: using the parameter action generation network to perform parameter action space mapping on the scene state vector set to obtain an initial camera parameter action set;

[0081] Specifically, the parameter action generation network can be trained using historical data. This network can predict the parameter adjustment action that should be performed under a given scene state by learning the association between the scene state vector and the camera parameter adjustment. The parameter action generation network is built and trained using the Python programming language in conjunction with the deep learning frameworks TensorFlow and Keras. The scene state vector set is taken as input, which contains the spatiotemporal characteristics of the target vehicle and the scene. The parameter action generation network maps the scene state vector to a continuous action space through a series of fully connected layers and activation functions. For example, if the camera parameters include focal length, exposure time, and ISO, the network will output a three-dimensional vector representing the adjustment amount of these three parameters. For each vector in the scene state vector set, the network generates a corresponding parameter adjustment action to form an initial camera parameter action set. Check the initial camera parameter action set to ensure that each action is reasonable and matches the corresponding scene state vector.

[0082] Step S34: using the state value evaluation network to evaluate the action value of the initial camera parameter action set to obtain a camera parameter action value score set;

[0083] Specifically, a state value evaluation network can be trained using reinforcement learning methods, such as Q-learning or policy gradient methods. This network learns to evaluate the value of performing specific parameter adjustment actions under a given scene state. The Python programming language is used in conjunction with the deep learning frameworks TensorFlow and Keras to build and train the state value evaluation network. The initial camera parameter action set and the corresponding scene state vector are provided as input to the state value evaluation network. The state value evaluation network calculates a value score for each parameter adjustment action through a series of fully connected layers and activation functions. This score reflects the expected effect of the action, such as improved image quality or target tracking accuracy. For each action in the initial camera parameter action set, the network generates a corresponding value score to form a camera parameter action value score set.

[0084] Step S35: performing a policy gradient update on the parameter adjustment strategy-evaluation network based on the camera parameter action value score set to obtain an optimized parameter adjustment strategy-evaluation network;

[0085] Specifically, the Python environment can be configured, and the deep learning framework PyTorch and its related dependent libraries can be installed. Set up a PyTorch project and create a script for defining and training the policy network. Define the structure of the parameter adjustment policy-evaluation network. This network is a deep neural network, comprising an input layer, multiple hidden layers, and an output layer. The input layer receives the scene state vector, and the output layer outputs the value score of each parameter adjustment action. The value score set obtained in step S34 is used as training data. These data include parameter adjustment actions and corresponding value scores, which are used to train the network to learn how to evaluate the value of actions. Define a loss function, usually using mean square error (MSE) or cross entropy loss, to measure the difference between the value score output by the network and the actual value score. Select an optimizer, such as Adam or RMSprop, to update the weights of the network. The optimizer adjusts the network parameters according to the gradient of the loss function to minimize the loss. In each training iteration, the output value score of the network is first calculated by forward propagation, and then the loss function is calculated. Next, the gradient of the loss relative to the network parameters is calculated by backpropagation. Using the calculated gradient and the preset learning rate, the optimizer updates the weights of the network. This process aims to increase the probability of the network outputting actions with high value scores. The training process is repeated multiple times, and the network parameters are updated in each iteration until the network converges or reaches the preset number of iterations. Through multiple iterations of training and gradient updates, the network gradually learns the strategy of selecting the best parameter adjustment action under different scene states, thereby obtaining an optimized parameter adjustment strategy - evaluation network.

[0086] Step S36: Parameter mapping is performed on the scene state vector set using the optimized parameter adjustment strategy-evaluation network to obtain a camera parameter adjustment strategy set;

[0087] Specifically, the Python programming language can be used in conjunction with the deep learning framework PyTorch, as well as a previously trained and optimized parameter adjustment strategy-evaluation network. The scene state vector set obtained in step S31 is input into the optimized parameter adjustment strategy-evaluation network. The network outputs a set of parameter adjustment actions based on each scene state vector, which are the camera parameter adjustments that the network considers to be optimal under the current scene state. For each vector in the scene state vector set, the network generates a corresponding parameter adjustment strategy to form a camera parameter adjustment strategy set. These parameter adjustment strategies are applied to the actual camera system to guide the camera to perform adaptive adjustments based on the scene state. The performance indicators of the monitoring camera, such as image clarity, target tracking accuracy, etc., ensure that the camera performance is improved after the parameter adjustment strategy is applied.

[0088] Step S37: Screen the camera parameter adjustment strategy set to obtain an initial parameter adjustment plan; perform predictive compensation on the initial parameter adjustment plan to obtain a real-time parameter adjustment strategy.

[0089] Specifically, please refer to the sub-steps of step S37 for the detailed implementation process of this embodiment.

[0090] The present invention captures the dynamic changes of the scene through spatiotemporal feature coding, thereby improving the accuracy of parameter adjustment. By constructing a dual network architecture, including a parameter action generation network and a state value evaluation network, the intelligence of strategy generation is enhanced, and the quality and applicability of the adjustment strategy are improved. The selection of parameter actions is optimized through parameter action space mapping, while the state value evaluation network improves the efficiency of parameter adjustment and quickly identifies the most effective adjustment action. Self-optimization of the strategy is achieved through policy gradient updating, making the parameter adjustment strategy more accurate and efficient. The real-time performance of parameter adjustment is enhanced through the optimized parameter adjustment strategy-evaluation network, ensuring that the camera parameters can quickly respond to environmental changes.

[0091] Preferably, step S37 includes the following steps:

[0092] Step S371: collecting historical adjustment effects of the camera parameter adjustment strategy set to obtain historical adjustment effect data;

[0093] Specifically, a database system, such as MySQL or MongoDB, can be used to store and retrieve historical data of camera parameter adjustments. These data include detailed information on each parameter adjustment, such as image quality indicators before and after adjustment, environmental conditions, adjustment strategies, etc. Records related to camera parameter adjustments are extracted from the database, including descriptions of adjustment strategies, objective quality indicators of adjusted images (such as clarity, contrast, noise level, etc.), and subjective scores (such as user satisfaction survey results). The collected data is organized to form structured historical adjustment effect data. This data includes a unique identifier for each adjustment strategy, adjustment parameters, environmental conditions, image quality indicators, and user scores.

[0094] Step S372: constructing a strategy priority scoring model based on historical adjustment effect data to obtain a camera strategy scoring model;

[0095] Specifically, the policy priority scoring model can be constructed using the Python programming language in conjunction with the machine learning library scikit-learn. Select features related to policy effects from historical adjustment effect data, such as adjustment parameters, environmental conditions, image quality indicators, and user ratings. Use regression analysis or classification algorithms (such as random forests, gradient boosting trees, etc.) to train the model. The input of the model is the selected features, and the output is the priority score of the policy. Evaluate the accuracy and generalization ability of the model through cross-validation and external validation. Use part of the data to train the model and another part of the data to test the predictive performance of the model. According to the verification results, adjust the model parameters and algorithms to optimize the predictive ability of the model.

[0096] Step S373: using the camera strategy scoring model to perform priority scoring and evaluation on the camera parameter adjustment strategy set to obtain a camera strategy priority sequence;

[0097] Specifically, each parameter adjustment strategy can be scored using a camera strategy scoring model. Use the Python programming language with the machine learning library scikit-learn to load and apply the scoring model. Take the camera parameter adjustment strategy set as input, which includes different parameter adjustment actions, such as exposure, focal length, etc. For each strategy in the strategy set, extract relevant features and score them using the scoring model. The model outputs a score that represents the priority of the strategy. Based on the scores given by the model, generate a camera strategy priority sequence, in which each strategy is sorted from high to low according to its score. Check the priority sequence to ensure that the scoring logic is reasonable and that high-scoring strategies correspond to better parameter adjustment actions as expected.

[0098] Step S374: sorting and screening the camera parameter adjustment strategy set according to the camera strategy priority sequence to obtain an initial parameter adjustment plan;

[0099] Specifically, the Python programming language can be used to sort and filter the policies. The policy set is sorted according to the camera policy priority sequence. The sorting can be in descending order to ensure that the policies with the highest scores are at the top. Based on business needs and resource constraints, a certain number of policies are filtered out from the sorted policy set as the initial parameter adjustment scheme. For example, only the top 10% of the policies are selected. The filtered policies constitute the initial parameter adjustment scheme, which are considered to have the greatest potential to improve camera performance.

[0100] Step S375: performing time series prediction on the initial parameter adjustment scheme to obtain a camera parameter change trend graph;

[0101] Specifically, the Python programming language can be used with time series analysis libraries such as statsmodels or time series prediction models in the machine learning library scikit-learn, such as ARIMA (autoregressive integrated moving average model) or LSTM (long short-term memory network). Collect historical camera parameter adjustment data, including adjustment time points, parameter values, etc., to form a time series data set. Based on historical data, train a time series prediction model. For example, use an LSTM network to learn the pattern of parameter changes over time and predict future parameter changes. Input the data in the initial parameter adjustment plan into the trained time series prediction model, and the model outputs the change trend of camera parameters in the future. Based on the prediction results of the model, generate a camera parameter change trend graph, which shows the prediction curve of different parameters over time. Verify the accuracy of the prediction by comparing with the actual camera parameter change data, and adjust the model parameters as needed to improve the prediction accuracy.

[0102] Step S376: Compensate and correct the initial parameter adjustment plan according to the camera parameter change trend diagram to obtain a real-time parameter adjustment strategy.

[0103] Specifically, the Python programming language can be used in combination with feedback control algorithms in control theory, such as the PID controller (proportional-integral-differential controller), to perform parameter corrections based on trend graphs. Analyze the camera parameter change trend graph to identify whether the predicted parameter changes deviate from the desired trajectory. Based on the analysis results of the trend graph, calculate the required compensation amount. For example, if the predicted exposure time deviates from the target value, calculate the exposure amount that needs to be increased or decreased. Based on the calculated compensation amount, make real-time adjustments to the parameters in the initial parameter adjustment scheme. These adjustments can be executed immediately or planned to be executed at a certain point in the future. The adjusted parameter set constitutes a real-time parameter adjustment strategy.

[0104] The present invention improves the accuracy of strategy selection by collecting historical adjustment effect data, so that the system can select the best camera parameter adjustment strategy based on actual effects rather than theoretical analysis alone. The objectivity of strategy evaluation is enhanced by the constructed camera strategy scoring model, and the pros and cons of strategies are evaluated based on actual effect data rather than subjective judgment. By using the scoring model to perform priority scoring and evaluation on the strategy set, the priority sorting of the strategy is optimized, and the parameter adjustment strategy with the greatest chance of producing the best effect can be quickly identified, thereby improving the efficiency of parameter adjustment. The application of time series prediction enhances the accuracy of prediction, and the changing trend of camera parameters can be predicted and adjusted in advance. By compensating and correcting the initial parameter adjustment plan according to the predicted trend graph, dynamic compensation correction is achieved, which can dynamically adapt to environmental changes and maintain the real-time and effectiveness of parameter adjustment.

[0105] Preferably, step S4 comprises the following steps:

[0106] Step S41: extracting key parameter control points of the real-time parameter adjustment strategy to obtain a camera parameter control sequence;

[0107] Specifically, the Python programming language can be used with data analysis libraries such as Pandas to process and analyze real-time parameter adjustment strategies. First, identify the key parameters that have the greatest impact on camera performance, such as focal length, exposure time, ISO sensitivity, etc. The selection of these parameters is based on camera performance indicators and expert experience. For each key parameter, control points are extracted from the real-time parameter adjustment strategy, which define the adjustment range and target value of the parameter. For example, for focal length, the control points include the trigger conditions and target distance for autofocus. Arrange the control points of each key parameter in chronological order to form a camera parameter control sequence. This sequence guides how the camera adjusts its parameters in different situations.

[0108] Step S42: performing three-dimensional parameter state space modeling on the camera to obtain a camera parameter space model;

[0109] Specifically, you can use MATLAB or Python with computer vision libraries such as OpenCV and 3D modeling tools such as Blender to build a 3D parameter state space model of the camera. Define the 3D parameters of the camera, including physical parameters (such as focal length, aperture size) and environmental parameters (such as light intensity, scene depth). Construct a 3D state space in which each dimension represents a parameter and each point represents a parameter combination state. For example, one dimension represents different values ​​of focal length and another dimension represents different values ​​of exposure time. Use a 3D modeling tool to create a virtual camera model and define the adjustment range and relationship of the parameters in the model. This model allows you to simulate the effects of parameter adjustments in a virtual environment.

[0110] Step S43: performing parameter coupling evaluation on the camera parameter control sequence to obtain a camera parameter correlation matrix;

[0111] Specifically, the Python programming language can be used with statistical analysis libraries such as NumPy and SciPy to evaluate the coupling between parameters. The historical adjustment data of key parameters in the camera parameter control sequence are extracted from the database, including but not limited to focal length, exposure time, ISO sensitivity, etc. Statistical methods such as the Pearson correlation coefficient or the Spearman rank correlation coefficient are used to evaluate the correlation between different parameters. Based on the results of the coupling analysis, a camera parameter association matrix is ​​constructed. In this matrix, each element represents the strength of the correlation between two parameters, the diagonal elements of the matrix represent the correlation of the parameters themselves (always 1), and the off-diagonal elements represent the correlation between different parameters.

[0112] Step S44: constructing a parameter linkage constraint equation based on the camera parameter association matrix to obtain a camera parameter linkage constraint model, and integrating the camera parameter space model with the camera parameter linkage constraint model to obtain a camera parameter combination model;

[0113] Specifically, data analysis software (such as Python's Pandas library) can be used to analyze the camera parameter association matrix and identify strong correlations and potential dependencies between parameters. For example, it is found that under low-light conditions, an increase in ISO sensitivity is usually accompanied by an adjustment in shutter speed. Collect physical laws and empirical rules related to camera parameter adjustment. These rules come from photography theory, camera technical manuals, or experience sharing by professional photographers. Use MATLAB or Python with the symbolic computing library SymPy to start building parameter linkage constraint equations. For example, there is a rule that "when ISO sensitivity increases, exposure time should be reduced to avoid overexposure." Convert this rule into a mathematical equation, such as ExposureTime = f(ISO), where f is a function defined based on empirical data. In SymPy, define symbolic variables to represent camera parameters, such as ISO, ExposureTime, etc., and use these symbolic variables to build constraint equations. Ensure that these equations can accurately describe the linkage relationship between parameters. Integrate all constructed constraint equations into a parameter linkage constraint model. This model can represent complex dependencies between parameters at the symbolic level. Integrate the parameter linkage constraint model with the camera parameter space model. This involves applying constraint equations to the parameter space model to ensure that all parameter adjustments satisfy physical constraints and coupling relationships. For example, the linkage between exposure time and ISO sensitivity is integrated into the parameter space model to ensure that when ISO is adjusted, the exposure time can be automatically adjusted accordingly. The integrated camera parameter combination model is verified by simulating different shooting scenes. Check whether the model can accurately guide parameter adjustments under different conditions to achieve the best shooting effect. This integration takes into account the physical constraints and coupling relationships of the parameters to generate a comprehensive camera parameter combination model.

[0114] Step S45: constructing a fuzzy rule base for the camera parameter combination model to obtain a fuzzy control rule set;

[0115] Specifically, the fuzzy rule base can be constructed using MATLAB's Fuzzy Logic Toolbox or Python's Skfuzzy library. First, define fuzzy variables related to camera parameters, such as exposure time, ISO sensitivity, and focal length, and define fuzzy sets for each variable, such as "low", "medium", and "high". Determine membership functions for each fuzzy set, which describe the degree to which each exact value belongs to the fuzzy set. For example, for ISO sensitivity, define a triangular membership function where 0-200 is "low", 200-400 is "medium", and 400-800 is "high". Fuzzy rules are formulated based on expert knowledge and historical data. For example, if the scene brightness is "low", the ISO sensitivity should be set to "high"; if the scene brightness is "high", the ISO sensitivity should be set to "low". All fuzzy rule sets are combined into a fuzzy control rule set, which will be used for subsequent fuzzy reasoning operations.

[0116] Step S46: performing fuzzy reasoning operation on the camera parameter combination model based on the fuzzy control rule set to obtain parameter linkage evaluation data;

[0117] Specifically, MATLAB's Fuzzy Logic Toolbox or Python's Skfuzzy library can be used to perform fuzzy reasoning operations. The specific parameter values ​​in the real-time parameter adjustment strategy are fuzzified, that is, the membership of these values ​​in each fuzzy set is determined. The fuzzified input values ​​are reasoned using a fuzzy control rule set. This process involves fuzzy logic operations, such as fuzzy AND, OR, and NOT, as well as the aggregation of fuzzy rules. The output of the reasoning operation is the fuzzy value of each parameter, which represents the recommended adjustment direction and degree of the parameter in the current scenario. In order to convert the fuzzy value into a specific parameter adjustment value, a defuzzification method, such as the centroid method or the maximum membership method, is applied to obtain a specific parameter adjustment evaluation value. The calculated evaluation values ​​of each parameter are integrated into parameter linkage evaluation data, which provide a quantitative basis for the specific adjustment of the camera parameters.

[0118] Step S47: Optimize and iterate the camera parameter control sequence according to the parameter linkage evaluation data to obtain an optimized camera parameter control sequence, and adjust the camera parameters according to the optimized camera parameter control sequence to obtain an optimized parameter adjustment scheme.

[0119] Specifically, the Python programming language can be used in conjunction with an optimization algorithm library such as SciPy, which provides a variety of optimization algorithms, such as gradient descent or genetic algorithm. The camera parameter control sequence is optimized based on the parameter linkage evaluation data. For example, the gradient descent method is used to minimize image quality evaluation indicators such as mean square error (MSE) or structural similarity index (SSIM). The parameter control sequence is adjusted iteratively, and the image quality evaluation index is used to evaluate the effect after each adjustment, and the control sequence is updated according to the evaluation results. The effectiveness of the optimization is verified by actually applying the optimized parameter control sequence to the camera and observing the changes in image quality. After multiple iterative optimizations, an optimized camera parameter control sequence is obtained, which can maximize image quality or meet other performance indicators. The camera is controlled in real time according to the optimized parameter control sequence, an optimized parameter control scheme is generated, and its effect is continuously monitored for further fine-tuning.

[0120] The present invention improves the accuracy of parameter control by extracting key parameter control points, so that the system can accurately identify and adjust the parameters that have the greatest impact on camera performance. The spatial cognition of parameter adjustment is enhanced by three-dimensional parameter state space modeling, making the relationship between parameters and the adjustment effect more intuitive and easy to manage. The synergy between parameters is optimized by parameter coupling evaluation and the construction of an association matrix, and the overall effect of parameter adjustment is improved. The efficiency and effect of parameter linkage are improved by the construction and integration of a parameter linkage constraint model, and parameter linkage adjustment can be performed efficiently. The intelligence of parameter adjustment is enhanced by the construction of a fuzzy rule base and fuzzy reasoning operations, and the flexibility and adaptability of parameter adjustment are improved. The rapid response of parameter adjustment is achieved by defuzzification processing, and the response speed of parameter adjustment is accelerated. The optimization iteration based on parameter linkage evaluation data improves the iterative optimization capability of parameter adjustment, and continuously optimizes the parameter control sequence.

[0121] Preferably, step S5 comprises the following steps:

[0122] Step S51: intelligently reconstructing the parameters of the camera according to the optimized parameter control scheme, and collecting monitoring images of the camera after parameter reconstruction to obtain real-time monitoring images;

[0123] Specifically, the camera's API or SDK can be used. These tools allow programmatic adjustment of camera parameters such as focal length, exposure, ISO, etc. According to the optimized parameter control scheme, the camera parameters are automatically adjusted by writing scripts or using control software. For example, if the scheme requires adjusting the exposure time to adapt to low-light environments, the camera's exposure time parameters will be set through the API. After the parameter adjustment is completed, the camera is started to collect real-time monitoring images. This can be achieved by triggering the camera's shutter or starting the continuous shooting mode. The collected image data is transmitted to the processing system through the camera's data interface. These data streams are processed in real time to ensure image continuity and stability.

[0124] Step S52: extracting quality features of the real-time monitoring image to obtain an image quality feature set;

[0125] Specifically, image processing libraries such as OpenCV can be used to extract image quality features. Perform quality assessment on real-time monitoring images and extract features such as clarity, contrast, and color saturation. For example, an edge detection algorithm is used to assess image clarity, and histogram analysis is used to assess contrast and color saturation. For each image, a series of quality feature values ​​are calculated. For example, clarity can be calculated by an edge response function (ERF) or a Laplace operator, and contrast can be assessed by the histogram distribution of the image. The quality feature values ​​of each image are integrated into an image quality feature set. This set includes multiple quality features of multiple images.

[0126] Step S53: inputting the image quality feature set into a preset image quality assessment network to perform image quality assessment to obtain image quality assessment data;

[0127] Specifically, a deep learning model, such as a convolutional neural network (CNN), can be used, which is trained to evaluate image quality. This model is trained using a large amount of labeled image data. The image quality feature set is used as input data and input into the image quality assessment network. These feature sets include features such as image clarity, contrast, and color saturation. The image quality assessment network processes the input feature set and outputs an evaluation result through the multi-layer structure of the network. This result is a score or a probability distribution that indicates the quality of the image. The image quality assessment data output by the network is a quantitative indicator, which can be a score from 0 to 5 or a percentage from 0 to 100, depending on the design and training method of the network. The accuracy of the network evaluation is verified by comparing the evaluation data output by the network with the results of manual evaluation. If the results of the network evaluation are consistent with the manual evaluation, it means that the network can accurately evaluate the image quality.

[0128] Step S54: performing evaluation index quantization mapping on the image quality evaluation data to obtain an image quality evaluation feature vector;

[0129] Specifically, the Python programming language can be used in conjunction with data processing libraries such as Pandas and NumPy to perform quantitative mapping of evaluation indicators. According to the preset quantization rules, the image quality evaluation data is mapped to a quantitative feature space. For example, if the evaluation data is a score from 0 to 5, it is mapped to an interval from 0 to 1. The quantized evaluation indicators are constructed into a feature vector, which contains multiple quantitative indicators of image quality, such as clarity score, contrast score, etc. The effectiveness of the quantitative mapping is verified by analyzing the changes in the feature vector. If the feature vector can accurately reflect the changing trend of image quality, it means that the quantitative mapping is successful.

[0130] Step S55: performing camera parameter sensitivity identification on the image quality assessment feature vector based on the real-time monitoring image to obtain a sensitive camera parameter set, and prioritizing the sensitive camera parameter set to obtain a camera parameter control optimization sequence;

[0131] Specifically, sensitivity analysis can be performed using the Python programming language in conjunction with statistical analysis libraries such as Statsmodels. First, the image quality assessment feature vector is associated with the camera parameters. This includes analyzing the degree of impact of each parameter change on the image quality index. Use regression analysis or machine learning models (such as random forests) to identify which parameters are most sensitive to image quality. The model will evaluate the impact of each parameter change on the image quality index and give a sensitivity score. Based on the sensitivity score, the sensitive camera parameter set is prioritized. The highest priority parameters will be considered for adjustment first because they have the greatest impact on image quality. A camera parameter adjustment optimization sequence is generated based on the priority ranking of the parameters. This sequence adjusts the camera parameters in order of priority to optimize image quality. The effectiveness of sensitivity identification and priority ranking is verified by simulating or actually adjusting the camera parameters and observing the changes in image quality.

[0132] Step S56: updating the parameters of the optimized parameter control scheme according to the camera parameter control optimization sequence to obtain the final camera configuration parameters.

[0133] Specifically, the camera's API or SDK can be used to dynamically update parameters. According to the camera parameter control optimization sequence, the camera parameters are automatically adjusted in priority order by writing scripts or using control software. Dynamically adjust the camera parameters, re-acquire images after each adjustment, and use the image quality assessment network to evaluate the quality of the new images. Update the optimization parameter control scheme based on the new image quality assessment data. This involves adjusting the value of the parameter or changing the order of parameter adjustment. After a series of dynamic adjustments and evaluations, the final camera configuration parameters are determined. These parameters meet the image quality requirements while also taking into account other factors in actual applications, such as response speed and system stability. The effectiveness of the final configuration parameters is verified by continuously monitoring the camera performance and image quality. If necessary, the parameters can be adjusted again based on new data.

[0134] The present invention improves the real-time performance and quality of image acquisition through intelligent parameter reconstruction, ensuring that the camera can quickly adjust parameters and acquire high-definition real-time monitoring images, thereby improving the response speed of the monitoring system. By extracting and evaluating quality features of real-time monitoring images, the image quality can be objectively quantified. By quantitatively mapping the evaluation indicators and generating the image quality evaluation feature vector, the image quality can be finely analyzed and the accuracy of the evaluation can be improved. By identifying and prioritizing the sensitivity of camera parameters based on the image quality evaluation feature vector, the parameters that have the greatest impact on image quality can be identified and adjusted preferentially, and the efficiency and effect of parameter regulation can be optimized. By dynamically iterating and optimizing the parameters, the camera configuration parameters are always kept in the best state.

[0135] Preferably, the present invention further provides a camera adaptive adjustment system, which is used to execute the camera adaptive adjustment method as described above, and the camera adaptive adjustment system comprises:

[0136] The environmental feature recognition module is used to obtain a multi-source environmental sensor data set and an initial environmental image; perform environmental feature recognition on the initial environmental image according to the multi-source environmental sensor data set to obtain a scene environmental feature map;

[0137] The target tracking module is used to perform semantic scene segmentation on the scene environment feature map to obtain a semantic scene macro feature map; perform temporal attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map; perform feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map;

[0138] The adjustment strategy generation module is used to adjust the camera parameters according to the target vehicle-scene motion semantic graph to obtain a camera parameter adjustment strategy set; screen the camera parameter adjustment strategy set to obtain an initial parameter adjustment plan; predict and compensate the initial parameter adjustment plan to obtain a real-time parameter adjustment strategy;

[0139] The feedback control adjustment module is used to perform multi-parameter linkage modeling on the camera based on the real-time parameter adjustment strategy to obtain the camera parameter combination model; perform fuzzy controller evaluation on the camera parameter combination model to obtain parameter linkage evaluation data; perform feedback control adjustment based on the parameter linkage evaluation data to obtain an optimized parameter control scheme;

[0140] The parameter updating module is used to perform intelligent parameter reconstruction and image acquisition of the camera according to the optimized parameter control scheme to obtain real-time monitoring images; perform image quality assessment on the real-time monitoring images to obtain image quality assessment data; and update the parameters of the optimized parameter control scheme according to the image quality assessment data to obtain the final camera configuration parameters.

[0141] Preferably, a computer-readable storage medium stores a camera adaptive adjustment method as described above which can be loaded and executed by a processor.

[0142] 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.

[0143] The above description 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 camera adaptive adjustment method, characterized in that: The following steps are involved: Step S1: Acquire a multi-source environmental sensing data set and an initial environmental image; perform environmental feature recognition on the initial environmental image according to the multi-source environmental sensing data set to obtain a scene environmental feature map; Step S2: performing semantic scene segmentation on the scene environment feature map to obtain a semantic scene macro feature map; Perform temporal attention target tracking on the scene environment feature map to obtain the target vehicle motion feature map; perform feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain the target vehicle-scene motion semantic map; Step S3: adjusting the camera parameters according to the target vehicle-scene motion semantic graph to obtain a camera parameter adjustment strategy set; screening the camera parameter adjustment strategy set to obtain an initial parameter adjustment scheme; predicting and compensating the initial parameter adjustment scheme to obtain a real-time parameter adjustment strategy; Step S4: Based on the real-time parameter adjustment strategy, multi-parameter linkage modeling is performed on the camera to obtain a camera parameter combination model; a fuzzy controller is evaluated on the camera parameter combination model to obtain parameter linkage evaluation data; Perform feedback control adjustment based on parameter linkage evaluation data to obtain an optimized parameter control scheme; Step S5: intelligently reconstruct the camera parameters and collect images according to the optimized parameter control scheme to obtain a real-time monitoring image; perform image quality assessment on the real-time monitoring image to obtain image quality assessment data; update the optimized parameter control scheme according to the image quality assessment data to obtain the final camera configuration parameters.

2. The camera adaptive adjustment method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting environmental parameters of the environment where the camera is located to obtain a multi-source environmental sensing data set; wherein the multi-source environmental sensing data set includes a light intensity data stream and a depth measurement data stream; Step S12: collecting a range image of the environment where the camera is located to obtain an initial environment image; Step S13: performing compensation illumination enhancement on the initial environment image according to the illumination intensity data stream to obtain an environment enhanced illumination feature map; Step S14: performing distance gradient calibration on the initial environment image according to the depth measurement data stream to obtain an environment calibration depth measurement map; Step S15: performing feature fusion on the environment enhancement illumination feature map and the environment calibration depth measurement map to obtain a scene environment feature map.

3. The camera adaptive adjustment method according to claim 2, characterized in that: Step S14 includes the following steps: Step S141: performing bilateral filtering on the initial environment image according to the depth measurement data stream to obtain a filtered environment image; Step S142: Calculate the depth gradient of the filtered environment image to obtain a spatial depth gradient map; Step S143: constructing a depth calibration transformation matrix according to the spatial depth gradient map to obtain an image depth calibration parameter set; Step S144: performing perspective transformation correction on the filtered environment image using the image depth calibration parameter set to obtain an environment depth correction image; Step S145: reconstructing the environment depth correction map in three dimensions to obtain an environment calibration depth measurement map.

4. The camera adaptive adjustment method according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing scene decomposition on the scene environment feature map to obtain an environment scene scale feature set; Step S22: Calculate the regional importance of the scene environment feature map based on the environment scene scale feature set to obtain an environment scene attention distribution map; Step S23: performing sliding window feature extraction on the environmental scene attention distribution map to obtain a scene local feature data set; Step S24: performing scene region segmentation on the scene environment feature map according to the environment scene attention distribution map and the scene local feature data set to obtain a semantic scene macro feature map; Step S25: generating a temporal feature sequence for the scene environment feature map to obtain a temporal feature set of a moving target; Step S26: Modeling the spatiotemporal attention mechanism for the moving target temporal feature set to obtain the target vehicle attention weight map; Step S27: locating the moving target time series feature set based on the target vehicle attention weight map to obtain a target vehicle motion feature map; Step S28: performing feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map.

5. The camera adaptive adjustment method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing spatiotemporal feature encoding on the target vehicle-scene motion semantic graph to obtain a scene state vector set; Step S32: constructing a dual network architecture for the scene state vector set to obtain a parameter adjustment strategy-evaluation network; wherein the parameter adjustment strategy-evaluation network includes a parameter action generation network and a state value evaluation network; Step S33: using the parameter action generation network to perform parameter action space mapping on the scene state vector set to obtain an initial camera parameter action set; Step S34: using the state value evaluation network to evaluate the action value of the initial camera parameter action set to obtain a camera parameter action value score set; Step S35: performing a policy gradient update on the parameter adjustment strategy-evaluation network based on the camera parameter action value score set to obtain an optimized parameter adjustment strategy-evaluation network; Step S36: Parameter mapping is performed on the scene state vector set using the optimized parameter adjustment strategy-evaluation network to obtain a camera parameter adjustment strategy set; Step S37: Screen the camera parameter adjustment strategy set to obtain an initial parameter adjustment plan; perform predictive compensation on the initial parameter adjustment plan to obtain a real-time parameter adjustment strategy.

6. The camera adaptive adjustment method according to claim 5, characterized in that: Step S37 includes the following steps: Step S371: collecting historical adjustment effects of the camera parameter adjustment strategy set to obtain historical adjustment effect data; Step S372: constructing a strategy priority scoring model based on historical adjustment effect data to obtain a camera strategy scoring model; Step S373: using the camera strategy scoring model to perform priority scoring and evaluation on the camera parameter adjustment strategy set to obtain a camera strategy priority sequence; Step S374: sorting and screening the camera parameter adjustment strategy set according to the camera strategy priority sequence to obtain an initial parameter adjustment plan; Step S375: performing time series prediction on the initial parameter adjustment scheme to obtain a camera parameter change trend graph; Step S376: Compensate and correct the initial parameter adjustment plan according to the camera parameter change trend diagram to obtain a real-time parameter adjustment strategy.

7. The camera adaptive adjustment method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: extracting key parameter control points of the real-time parameter adjustment strategy to obtain a camera parameter control sequence; Step S42: performing three-dimensional parameter state space modeling on the camera to obtain a camera parameter space model; Step S43: performing parameter coupling evaluation on the camera parameter control sequence to obtain a camera parameter correlation matrix; Step S44: constructing a parameter linkage constraint equation based on the camera parameter association matrix to obtain a camera parameter linkage constraint model, and integrating the camera parameter space model with the camera parameter linkage constraint model to obtain a camera parameter combination model; Step S45: constructing a fuzzy rule base for the camera parameter combination model to obtain a fuzzy control rule set; Step S46: performing fuzzy reasoning operation on the camera parameter combination model based on the fuzzy control rule set to obtain parameter linkage evaluation data; Step S47: Optimize and iterate the camera parameter control sequence according to the parameter linkage evaluation data to obtain an optimized camera parameter control sequence, and adjust the camera parameters according to the optimized camera parameter control sequence to obtain an optimized parameter adjustment scheme.

8. The camera adaptive adjustment method according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: intelligently reconstructing the parameters of the camera according to the optimized parameter control scheme, and collecting monitoring images of the camera after parameter reconstruction to obtain real-time monitoring images; Step S52: extracting quality features of the real-time monitoring image to obtain an image quality feature set; Step S53: inputting the image quality feature set into a preset image quality assessment network to perform image quality assessment to obtain image quality assessment data; Step S54: performing evaluation index quantization mapping on the image quality evaluation data to obtain an image quality evaluation feature vector; Step S55: performing camera parameter sensitivity identification on the image quality assessment feature vector based on the real-time monitoring image to obtain a sensitive camera parameter set, and prioritizing the sensitive camera parameter set to obtain a camera parameter control optimization sequence; Step S56: updating the parameters of the optimized parameter control scheme according to the camera parameter control optimization sequence to obtain the final camera configuration parameters.

9. A camera adaptive adjustment system, characterized in that: Used to execute the camera adaptive adjustment method as claimed in claim 1, the camera adaptive adjustment system comprises: The environmental feature recognition module is used to obtain a multi-source environmental sensor data set and an initial environmental image; perform environmental feature recognition on the initial environmental image according to the multi-source environmental sensor data set to obtain a scene environmental feature map; The target tracking module is used to perform semantic scene segmentation on the scene environment feature map to obtain a semantic scene macro feature map; perform temporal attention target tracking on the scene environment feature map to obtain a target vehicle motion feature map; perform feature constraint fusion on the semantic scene macro feature map and the target vehicle motion feature map to obtain a target vehicle-scene motion semantic map; The adjustment strategy generation module is used to adjust the camera parameters according to the target vehicle-scene motion semantic graph to obtain a camera parameter adjustment strategy set; screen the camera parameter adjustment strategy set to obtain an initial parameter adjustment plan; predict and compensate the initial parameter adjustment plan to obtain a real-time parameter adjustment strategy; The feedback control adjustment module is used to perform multi-parameter linkage modeling on the camera based on the real-time parameter adjustment strategy to obtain the camera parameter combination model; perform fuzzy controller evaluation on the camera parameter combination model to obtain parameter linkage evaluation data; perform feedback control adjustment based on the parameter linkage evaluation data to obtain an optimized parameter control scheme; The parameter updating module is used to perform intelligent parameter reconstruction and image acquisition of the camera according to the optimized parameter control scheme to obtain real-time monitoring images; perform image quality assessment on the real-time monitoring images to obtain image quality assessment data; and update the parameters of the optimized parameter control scheme according to the image quality assessment data to obtain the final camera configuration parameters.

10. A computer-readable storage medium storing a camera adaptive adjustment method as claimed in any one of claims 1 to 8 that can be loaded and executed by a processor.

Images