Intelligent image calibration system based on vehicle-mounted AI

Abstract

The invention provides an intelligent image calibration system based on a vehicle-mounted AI, and the system comprises a vehicle-mounted image collection device, and also comprises an image marking subsystem and an image calibration subsystem. The image marking subsystem comprises an image screening module, an image processing module, a feature marking module and a feature construction module; the image calibration subsystem comprises a feature index module, an extraction marking module, a feature comparison module and a parameter correction module; element features are determined from environment image information of a scene point through an image marking subsystem, an element feature model is configured in advance, when a vehicle passes through a scenic spot again, an image calibration subsystem compares reference shooting features with actual shooting features to generate comparison difference information, then image parameters and position parameters at each moment are calibrated, and the scene point is obtained. And the image acquisition unit is configured in an optimal acquisition state, so that the reliability of vehicle image acquisition is improved, and support is provided for functions of intelligent driving, obstacle avoidance and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of image calibration, and particularly to an intelligent image calibration system based on in-vehicle AI. Background Art

[0002] With the rapid development of automotive intelligent technologies, in-vehicle AI systems are playing an increasingly important role in functions such as vehicle autonomous driving, assisted driving, and safety hazard avoidance. Among them, the in-vehicle image acquisition and processing system, as a key component for a vehicle to perceive the surrounding environment, its performance directly affects the accuracy and reliability of functions such as intelligent driving and obstacle avoidance. In recent years, significant progress has been made in image recognition technologies based on deep learning, enabling vehicles to identify various types of objects and scenes in real time. However, due to the complex and variable driving environment and the change of lighting conditions, how to ensure that the image acquisition system can obtain high-quality image data in different scenarios remains a technical problem to be solved urgently.

[0003] Most of the existing in-vehicle image acquisition units rely on fixed parameter configurations and lack the ability to adapt to real-time scene changes. For example, when driving at night, the image acquisition unit with fixed parameters often fails to effectively capture clear image details, resulting in blurred boundaries of feature objects, which in turn affects subsequent image analysis and decision-making; while the existing self-calibration methods can usually only compensate for the degradation of image quality caused by factors such as lens distortion to a certain extent, and cannot effectively optimize and adjust for complex dynamic scenes. Summary of the Invention

[0004] In view of this, the problem to be solved by the present invention is to provide an intelligent image calibration system based on in-vehicle AI.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is:

[0006] An intelligent image calibration system based on in-vehicle AI includes an in-vehicle image acquisition device for acquiring environmental image information, and also includes an image marking subsystem and an image calibration subsystem;

[0007] The image marking subsystem includes: an image screening module, an image processing module, a feature marking module, and a feature construction module;

[0008] The image screening module is configured with a preset tile screening strategy to screen the environmental image information through the tile screening strategy to determine the environmental image frame as the image to be calibrated;

[0009] The image processing module is configured with feature element screening conditions and determines element features from the image to be calibrated through the feature element screening conditions;

[0010] The feature marking module is configured with a feature element calculation algorithm, which is used to calculate the cumulative calibration reliability value of the element feature. When the cumulative calibration reliability value is greater than the preset value, the element feature is calibrated;

[0011] The feature construction module obtains the marked element features to construct an element feature model, and generates a pose index of the element feature model according to the vehicle position information;

[0012] The image calibration subsystem includes: a feature index module, an extraction and marking module, a feature comparison module, and a parameter correction module;

[0013] The feature index module is configured with a preset feature comparison strategy, which obtains the corresponding environmental image information in real time and identifies the corresponding element features through comparison with the element feature database;

[0014] The extraction and marking module determines the relative shooting pose of the vehicle according to the relationship between the element feature and the element feature model, and generates an optimal shooting moment according to the relative shooting pose and the vehicle driving path information;

[0015] The feature comparison module extracts the environmental image frame at the optimal shooting moment and extracts the corresponding element feature as the actual shooting feature, extracts the element feature at the corresponding shooting pose from the element feature model as the reference shooting feature, and compares the reference shooting feature and the actual shooting feature through a preset feature comparison algorithm to generate comparison difference information; The parameter correction module retrieves the corresponding correction instruction from the preset correction database according to the comparison difference information, and configures the corresponding image acquisition device through the correction instruction.

[0016] The tile screening strategy includes extracting the environmental image frames in the corresponding environmental image information at a preset time interval, calculating the environmental clarity of each environmental image frame through a preset clarity calculation algorithm, and screening the environmental image frames with environmental clarity greater than the preset environmental clarity benchmark;

[0017] Calculate the feature abundance value of the environmental image frame through a preset abundance calculation algorithm, and use the environmental image frame with the largest feature abundance value as the environmental image frame to be calibrated.

[0018] The expression of the clarity calculation algorithm is:

[0019]

[0020] G x (x,y) is the gradient amplitude in the x direction, G y (x,y) is the gradient amplitude in the y direction, and C(x,y) represents the environmental clarity.

[0021] The expression of the abundance calculation algorithm is:

[0022]

[0023] H is the entropy of the image, n i is the number of occurrences of gray level i, and N is the total number of pixels.

[0024] The extraction and marking module places the element feature model in a reference coordinate system, and determines a number of virtual shooting points in the reference coordinate system, each virtual shooting point corresponds to a shooting feature, determines the shooting point coordinates corresponding to the shooting feature closest to the element feature, and maps the corresponding relative shooting posture according to the shooting point coordinates.

[0025] The feature element screening conditions include a boundary difference sub-condition, an internal abundance sub-condition and an identifiable sub-condition; when the boundary difference value of a certain image feature is greater than a preset boundary reference value, it is regarded as an element feature.

[0026] The expression of the calculation algorithm of the boundary difference value is:

[0027] F(x,y)=w q ·B(x,y)+w2·I(x,y)+w3·D(x,y)

[0028] F(x,y) is the boundary difference value of the image at the coordinate (x,y), B(x,y) is the boundary difference sub-condition, I(x,y) is the internal abundance sub-condition, D(x,y) is the identifiability sub-condition, and w1, w2, and w3 are weight coefficients.

[0029] The expression of the element feature calculation algorithm is:

[0030]

[0031] R(f) is the cumulative calibrated reliability value of feature f, T is the length of the observation time window, t is the time variable, α is the attenuation factor, F(t) is the clarity of the environmental image collected at time t, and G(f,t) is the abundance of feature f collected at time t.

[0032] The construction of the feature feature model includes constructing an initial feature feature model by splicing the marked feature features, and completing the boundary part of the initial feature feature model with AI, obtaining the index image through the completed feature feature model, inferring the feature features of the index image, and then splicing it to form the ultimate feature feature model.

[0033] By means of a preset feature comparison algorithm, calculate the sharpness difference and the exposure difference, and generate a difference feature vector based on the sharpness difference and the spreading difference. The difference information includes the difference feature vector, the ambient illuminance value, and the shooting time value.

[0034] The advantages and positive effects of the present invention are as follows:

[0035] The element features are determined from the environmental image information of the scene point by the image marking subsystem, and the element feature model is pre-configured. When the vehicle passes through the scene point again, the image calibration subsystem determines the shooting point coordinates corresponding to the shooting feature closest to the element feature, and maps the corresponding relative shooting pose according to the shooting point coordinates, thereby generating the optimal shooting time. At the optimal shooting time, an environmental image frame is extracted and the corresponding element features are extracted as the actual shooting features, and the element features corresponding to the shooting pose are extracted from the element feature model as the reference shooting features. The reference shooting features and the actual shooting features are compared by a preset feature comparison algorithm to generate comparison difference information, thereby calibrating the image parameters and position parameters at each moment, configuring the image acquisition unit in the optimal acquisition state, improving the reliability of vehicle image acquisition, and providing support for functions such as intelligent driving and obstacle avoidance. Description of the Drawings

[0036] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention.

[0037] In the drawings:

[0038] Figure 1 is a flowchart of the intelligent image calibration system based on in-vehicle AI of the present invention. Detailed Embodiments

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] It should be noted that when a component is referred to as "fixed to" another component, it can be directly on the other component or there may be an intermediate component. When a component is considered to be "connected to" another component, it can be directly connected to the other component or there may be an intermediate component at the same time. When a component is considered to be "disposed on" another component, it can be directly disposed on the other component or there may be an intermediate component at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used in this article are only for the purpose of illustration.

[0041] Unless otherwise defined, all technical and scientific terms used in this article have the same meaning as commonly understood by those skilled in the technical field to which this invention belongs. The terms used in the description of the present invention in this article are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used in this article includes any and all combinations of one or more of the related listed items.

[0042] As Figure 1 shown, the present invention provides an intelligent image calibration system based on in-vehicle AI, including an in-vehicle image acquisition device. The in-vehicle image acquisition device includes a camera type for acquiring environmental image information. The environmental image information is visual data about the environment around the vehicle, and also includes an image marking subsystem and an image calibration subsystem.

[0043] The image marking subsystem includes: an image screening module, an image processing module, a feature marking module, and a feature construction module;

[0044] The image screening module is configured with a preset tile screening strategy. The tile screening strategy includes extracting environmental image frames corresponding to the environmental image information at a preset time interval. An environmental image frame refers to a single image intercepted from the environmental image information (such as a video stream), and calculating the environmental clarity of each environmental image frame through a preset clarity calculation algorithm, screening the environmental image frames with an environmental clarity greater than a preset environmental clarity benchmark, and then calculating the feature abundance value of the environmental image frame through a preset abundance calculation algorithm. The feature abundance value reflects the richness and obviousness of the features in the image. The larger the feature abundance value, the more feature information the frame image contains, such as corner points, edges, textures, etc. These rich feature information helps to improve the accuracy of camera calibration because a large number of clear and accurate feature points are required during the calibration process to estimate the parameters of the camera. Therefore, the environmental image frame with the largest feature abundance value is used as the environmental image frame to be calibrated;

[0045] The image processing module is configured with feature element screening conditions, and determines element features from the image to be calibrated through the feature element screening conditions. The feature element screening conditions include a boundary difference sub-condition, an internal abundance sub-condition, and an identifiable sub-condition; when the boundary difference value of a certain image feature is greater than a preset boundary reference value, it indicates that the feature has a unique visual performance on the boundary, is easy to distinguish from other features, and is more likely to be accurately identified and extracted in subsequent processing and analysis, thereby improving the accuracy and efficiency of processing. Then, this image feature is used as the determined element feature;

[0046] The feature marking module is configured with a feature element calculation algorithm, and the feature element calculation algorithm is used to calculate the cumulative calibration reliability value of the element feature. When the cumulative calibration reliability value is greater than a preset value, mark the element feature;

[0047] The cumulative calibration reliability value is the result of comprehensively calculating the performance of the element feature in multiple calibration processes. By calculating the cumulative calibration reliability value, the stability and accuracy of the feature in different frames and different scenarios can be comprehensively considered, so as to more comprehensively evaluate the quality of the feature, and further improve the accuracy and stability of calibration;

[0048] The feature construction module obtains the marked element features to construct an element feature model, and generates a pose index of the element feature model according to the vehicle position information, and associates the generated pose index with the element feature model, so as to quickly locate and access the element feature model in the corresponding pose during vehicle driving, which can be achieved by establishing an index mapping table or using a database management system;

[0049] The construction of the element feature model includes splicing the marked element features to construct an initial element feature model, and using AI to complement the boundary part of the initial element feature model. After obtaining the index image through the complemented element feature model, and estimating the element features of the index image, then splicing is performed to form the ultimate element feature model.

[0050] The image calibration subsystem includes: a feature index module, an extraction and marking module, a feature comparison module, and a parameter correction module;

[0051] The feature index module is configured with a preset feature comparison strategy. The feature comparison strategy obtains the corresponding environmental image information in real time, and identifies the corresponding element features through comparison with the element feature database;

[0052] The extraction and marking module places the element feature model in the reference coordinate system, and determines several virtual shooting points in the reference coordinate system. Each virtual shooting point corresponds to a shooting feature. Determine the shooting point coordinates corresponding to the shooting feature closest to the element feature, and map the corresponding relative shooting pose according to the shooting point coordinates, and then generate the optimal shooting moment;

[0053] Determining the shooting point coordinates corresponding to the shooting feature closest to the feature feature can be accomplished by shape matching. Extract the shape contour of the feature feature model. Usually, an edge detection algorithm (such as the Canny operator) can be used to obtain the boundary information of the feature. Then, perform shape analysis on the shooting feature corresponding to each virtual shooting point. For example, if the feature feature is a rectangular object, then the similarity between the contour of the shooting feature and the contour of the rectangular feature feature can be calculated. By comparing geometric parameters such as the perimeter, area, and ratio of corresponding side lengths of the two, find the shooting feature that is most similar in shape. For features with simple regular shapes, such as circles and triangles, the difference between the key dimensions of the shooting feature and the feature feature (such as the radius of a circle and the side length of a triangle) can be calculated using geometric formulas. The shooting point coordinates corresponding to the shooting feature with the smallest difference are the shooting point coordinates corresponding to the shooting feature closest to the feature feature;

[0054] The feature comparison module extracts the environmental image frame at the optimal shooting moment and extracts the corresponding feature feature as the actual shooting feature, extracts the feature feature corresponding to the shooting pose from the feature feature model as the reference shooting feature, and compares the reference shooting feature and the actual shooting feature through a preset feature comparison algorithm to generate comparison difference information;

[0055] For example, calculate the sharpness difference and the exposure difference, and generate a difference feature vector based on the sharpness difference and the exposure difference. The difference information includes the difference feature vector, the environmental illuminance value, and the shooting moment value. The parameter correction module retrieves the corresponding correction instruction from a preset correction database according to the comparison difference information. The correction database stores correction strategies and instructions for different types of differences. For example, if the sharpness difference indicates that the image is blurred, then there may be correction instructions in the correction database for adjusting the lens focus or the anti-shake function; if the exposure difference shows that the image is too dark or too bright, there may be instructions for adjusting the aperture size or the shutter speed. If the correction instruction indicates that the sharpness needs to be improved, the control system of the image acquisition device will adjust the focusing mechanism of the lens according to the instruction, so that the lens quickly and accurately focuses on the target object, thereby improving the sharpness of the image. If the exposure needs to be adjusted, the device will correspondingly change the aperture size or the shutter speed to obtain an appropriate exposure value, and then configure the corresponding image acquisition device through the correction instruction.

[0056] Specifically, the expression of the sharpness calculation algorithm is:

[0057]

[0058] C(x,y) is the environmental sharpness, G x (x,y) is the gradient amplitude in the x direction, G y(x, y) is the gradient magnitude in the y direction;

[0059] G x (x, y) and G y (x, y) is calculated by the Sobel operator as follows:

[0060]

[0061] The clarity calculation algorithm is used to evaluate the environmental clarity of each environmental image frame. In this application, the clarity of the image is measured based on the gradient information of the image because the gradient can reflect the sharpness of the edges in the image. The Sobel operator is used to calculate the gradient magnitudes in the horizontal and vertical directions, and then these two gradient magnitudes are combined to obtain the final clarity metric.

[0062] Specifically, the expression of the abundance calculation algorithm is:

[0063]

[0064] H is the entropy of the image, n i is the number of occurrences of the gray level i, N is the total number of pixels, and p(i) is the probability density of the pixel value being i. When calculating the probability density p(i) of the image, the interval 0 - 255 is selected because the gray level range of the image is usually 0 - 255. Such a selection facilitates normalization processing, enabling the probability density function to accurately reflect the distribution of each gray level in the image;

[0065] The feature abundance value can be understood as the amount of useful information contained in the image, such as the number of features like edges and corners. In this application, the entropy of the image is used as a measure of feature abundance because entropy can reflect the richness of information in the image.

[0066] Specifically, the expression of the boundary difference value calculation algorithm is:

[0067] F(x, y) = w1·B(x, y) + w2·I(x, y) + w3·D(x, y)

[0068] F(x, y) is the boundary difference value of the image at the coordinate (x, y), used to measure whether the image features meet the conditions of the element features. B(x, y) is the boundary difference sub - condition, reflecting the edge strength of the image features. I(x, y) is the internal abundance sub - condition, indicating the richness inside the image features. D(x, y) is the discriminability sub - condition, indicating the uniqueness and distinguishability of the image features. w1, w2, and w3 are weight coefficients used to adjust the importance of each sub - condition;

[0069] Specifically, the expression of the element feature calculation algorithm is:

[0070]

[0071] R(f) is the cumulative calibration reliability value of the feature f, T is the length of the observation time window, t is the time variable, α is the attenuation factor used to control the influence degree of historical data on the current result, making the recent data have a greater weight than the far - term data, F(t) is the clarity of the environmental image collected at time t, which is calculated by the clarity calculation algorithm, and G(f,t) is the abundance of the feature f collected at time t, which is calculated by the abundance calculation algorithm.

[0072] The working principle and process of the present invention are as follows:

[0073] In the image marking subsystem, the image screening module is configured with a preset tile screening strategy to screen the environmental image information through the tile screening strategy to determine the environmental image frame as the image to be calibrated; the image processing module is configured with feature element screening conditions and determines the feature elements from the image to be calibrated through the feature element screening conditions; the feature marking module is configured with a feature element calculation algorithm, and the feature element calculation algorithm is used to calculate the cumulative calibration reliability value of the feature element. When the cumulative calibration reliability value is greater than the preset value, the feature element is calibrated; the feature construction module obtains the marked feature elements to construct a feature element model, generates a pose index of the feature element model according to the vehicle position information, and associates the generated pose index with the feature element model to quickly locate and access the feature element models in different poses as reference models;

[0074] In the image calibration subsystem, the feature indexing module is configured with a preset feature comparison strategy. The feature comparison strategy obtains the corresponding environmental image information in real time and identifies the corresponding feature features through comparison with the feature feature database; the extraction and marking module places the feature feature model in the reference coordinate system and determines a number of virtual shooting points in the reference coordinate system. Each virtual shooting point corresponds to a shooting feature. The shooting point coordinates corresponding to the shooting feature closest to the feature feature are determined, and the corresponding relative shooting pose is mapped according to the shooting point coordinates, thereby generating the optimal shooting moment; the feature comparison module includes extracting the environmental image frame at the optimal shooting moment and extracting the corresponding feature features as the actual shooting features, extracting the feature features corresponding to the shooting pose from the feature feature model as the reference shooting features, and comparing the reference shooting features and the actual shooting features through a preset feature comparison algorithm to generate comparison difference information; the parameter correction module retrieves the corresponding correction instruction from the preset correction database according to the comparison difference information, and configures the corresponding image acquisition device through the retrieved correction instruction. The correction database stores correction strategies and instructions for different types of differences. For example, if the sharpness difference indicates that the image is blurred, then there may be correction instructions in the correction database for adjusting the lens focal length or anti-shake function in the correction database; if the exposure difference shows that the image is too dark or too bright, there may be instructions for adjusting the aperture size or shutter speed. If the correction instruction indicates that the sharpness needs to be improved, the control system of the image acquisition device will adjust the focusing mechanism of the lens according to the instruction, so that the lens can quickly and accurately focus on the target object, thereby improving the sharpness of the image. If the exposure needs to be adjusted, the device will change the aperture size or shutter speed accordingly to obtain a suitable exposure value.

[0075] The above has described the embodiments of the present invention in detail, but the content described is only the preferred embodiments of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope covered by this patent.

Claims

1. An intelligent image calibration system based on in-vehicle AI, characterized in that, It includes an in-vehicle image acquisition device for acquiring environmental image information, and also includes an image marking subsystem and an image calibration subsystem; The image marking subsystem includes: an image screening module, an image processing module, a feature marking module, and a feature construction module; The image screening module is configured with a preset tile screening strategy, and screens the environmental image information through the tile screening strategy to determine the environmental image frame to be calibrated; The image processing module is configured with feature element screening conditions, and determines element features from the image to be calibrated through the feature element screening conditions; The feature marking module is configured with a feature element calculation algorithm, and the feature element calculation algorithm is used to calculate the cumulative calibration reliability value of the element feature. When the cumulative calibration reliability value is greater than a preset value, the element feature is calibrated; The feature construction module obtains the marked element features to construct an element feature model, and generates a pose index of the element feature model according to the vehicle position information; The image calibration subsystem includes: a feature index module, an extraction marking module, a feature comparison module, and a parameter correction module; The feature index module is configured with a preset feature comparison strategy, and the feature comparison strategy obtains the corresponding environmental image information in real time, and identifies the corresponding element features through comparison with the element feature database; The extraction marking module determines the relative shooting pose of the vehicle according to the relationship between the element feature and the element feature model, and generates an optimal shooting moment according to the relative shooting pose and the vehicle driving path information; The feature comparison module extracts the environmental image frame at the optimal shooting moment and extracts the corresponding element features as the actual shooting features at the optimal shooting moment, extracts the element features corresponding to the shooting pose from the element feature model as the reference shooting features, and compares the reference shooting features and the actual shooting features through a preset feature comparison algorithm to generate comparison difference information; The parameter correction module retrieves the corresponding correction instruction from the preset correction database according to the comparison difference information, and configures the corresponding image acquisition device through the correction instruction; 2. The intelligent image calibration system based on in-vehicle AI according to claim 1, wherein The tile screening strategy includes extracting the environmental image frames in the corresponding environmental image information at a preset time interval, and calculating the environmental clarity of each environmental image frame through a preset clarity calculation algorithm, and screening the environmental image frames with the environmental clarity greater than the preset environmental clarity benchmark; Calculate the feature abundance value of the environmental image frame through a preset abundance calculation algorithm, and use the environmental image frame with the largest feature abundance value as the environmental image frame to be calibrated; 3. The intelligent image calibration system based on in-vehicle AI according to claim 2, wherein, The expression of the clarity calculation algorithm is: G x (x, y) is the gradient magnitude in the x direction, G y (x, y) is the gradient magnitude in the y direction, and C(x, y) represents the environmental clarity.

4. The intelligent image calibration system based on in-vehicle AI according to claim 2, characterized in that, The expression of the abundance calculation algorithm is: H is the entropy of the image, and n i is the occurrence times of the gray level i, and N is the total number of pixels.

5. The intelligent image calibration system based on in-vehicle AI according to claim 1, wherein, The extraction marking module places the element feature model in a reference coordinate system, determines a number of virtual shooting points in the reference coordinate system, each virtual shooting point corresponds to a shooting feature, determines the shooting point coordinates corresponding to the shooting feature closest to the element feature, and maps the corresponding relative shooting pose according to the shooting point coordinates; 6. The intelligent image calibration system based on in-vehicle AI according to claim 1, characterized in that, The feature element screening conditions include a boundary difference sub-condition, an internal abundance sub-condition, and an identifiable sub-condition; when the boundary difference value of a certain image feature is greater than a preset boundary reference value, it is used as an element feature; 7. The intelligent image calibration system based on in-vehicle AI according to claim 6, characterized in that, The expression of the calculation algorithm of the boundary difference value is: F(x,y) = w1·B(x,y) + w2·I(x,y) + w3·D(x,y) F(x,y) is the boundary difference value of the image at coordinates (x,y), B(x,y) is the boundary difference sub-condition, I(x,y) is the internal abundance sub-condition, D(x,y) is the recognizable sub-condition, and w1, w2, and w3 are weight coefficients.

8. The intelligent image calibration system based on in-vehicle AI according to claim 1, wherein, The expression of the element feature calculation algorithm is as follows: R(f) is the cumulative calibration reliability value of the element feature f, T is the length of the observation time window, t is the time variable, α is the decay factor, F(t) is the clarity of the environmental image collected at time t, and G(f,t) is the abundance of the element feature f collected at time t.

9. The intelligent image calibration system based on in-vehicle AI according to claim 1, characterized in that The construction of the element feature model includes splicing the labeled element features to construct an initial element feature model, complementing the boundary part of the initial element feature model in combination with AI, obtaining an index image through the complemented element feature model, and then splicing after inferring the element features of the index image to form an ultimate element feature model.

10. The intelligent image calibration system based on in-vehicle AI according to claim 1, characterized in that, Through a preset feature comparison algorithm, the clarity difference and the exposure difference are calculated, and a difference feature vector is generated according to the clarity difference and the spread difference. The difference information includes the difference feature vector, the environmental illumination value, and the shooting time value.

Images