Image acquisition control method, device and equipment based on machine vision

Abstract

The invention provides an image acquisition control method, device and equipment based on machine vision. The method comprises the following steps: selecting a high-frame-rate image sensor acquisition mode according to environment state parameters; adjusting exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; fusing the optical flow speed and the brightness of the image flow into a weighting factor according to the exposure parameters, and injecting the weighting factor into an image space to generate a scene partition mask; and the scene partition mask is aligned with the exposure images with different exposure parameters, and the exposure images are fused into a target acquisition image. Through the implementation of the scheme of the invention, the environment state parameters are generated in real time, the corresponding acquisition mode is selected, the exposure parameters are optimized, the optical flow and the brightness data are fused to generate the weighting factor, the scene partition mask is created, the multi-exposure image is aligned through the scene partition mask, the optimal exposure part of each area is accurately fused, and the high-quality target image is obtained. Therefore, the image acquisition efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of image processing, and in particular, to an image acquisition control method, apparatus, and device based on machine vision. Background Art

[0002] With the rapid development of machine vision technology, image acquisition based on machine vision has been widely used in industrial inspection, medical diagnosis, intelligent monitoring and other fields. However, in actual application scenarios, the image acquisition process often faces the dual challenges of motion blur and high dynamic range scenes, which seriously affect the image quality and the accuracy of subsequent processing.

[0003] During the image capture process, when the object being photographed or the camera itself is in a high-speed motion state, if an inappropriate image acquisition time control strategy is adopted, it is easy to cause motion blur in the image. This blurring effect will make the image edges unclear and the detail information lost, thus reducing the accuracy of image processing and analysis. Traditional exposure time control methods usually use fixed exposure parameters and cannot be intelligently adjusted according to the motion state of the target object, making it difficult to effectively reduce motion blur while ensuring the image brightness. On the other hand, in high dynamic range scenes, such as an environment where strong light and shadow coexist, conventional image acquisition techniques often have difficulty capturing the detail information of both bright and dark areas at the same time. Existing image acquisition trigger mechanisms are usually based on preset time intervals or simple environmental parameters and lack the ability to accurately perceive and adapt to the scene illumination conditions. This results in either overexposure of the bright part or loss of dark part details in complex lighting environments, and it is impossible to obtain high-quality panoramic images. Summary of the Invention

[0004] This application provides an image acquisition control method, apparatus, and device based on machine vision, which is used to solve the problem that high-quality panoramic images cannot be obtained with simply set exposure parameters and environmental parameters in related technologies.

[0005] In the first aspect of this application, an image acquisition control method based on machine vision is provided. The image acquisition control method based on machine vision includes: Continuously acquiring an image stream through a high-frame-rate image sensor and determining corresponding environmental state parameters; Selecting a high-frame-rate image sensor acquisition mode according to the environmental state parameters; Adjusting the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; Fusing the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and injecting the weighting factor into the image space to generate a scene partition mask; Fusing the exposure images into a target acquisition image by aligning the scene partition mask with the exposure images of different exposure parameters.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, the step of continuously collecting an image stream through a high-frame-rate image sensor and determining corresponding environmental state parameters includes: Continuously collect an image stream through a high-frame-rate image sensor; Generate the motion intensity parameter by normalizing the pixel displacement field of the image stream; Extract the luminance component of the image stream, determine the luminance extreme value by fitting a bimodal histogram, and perform logarithmic quantization based on the ratio of the luminance extreme values to generate the dynamic range parameter; Fuse the motion intensity parameter and the dynamic range parameter to generate an environmental state parameter.

[0007] Optionally, in the second implementation manner of the first aspect of the present application, the step of selecting the acquisition mode of the high-frame-rate image sensor according to the environmental state parameter includes: Perform Gaussian membership calculation on the motion intensity parameter to generate a first fuzzy factor, and at the same time perform S-type membership calculation on the dynamic range parameter to generate a second fuzzy factor; Cross-verify the first fuzzy factor and the second fuzzy factor through a fuzzy rule base; When the first fuzzy factor reaches the strong motion membership interval and the second fuzzy factor belongs to the low dynamic range membership interval, select the motion-priority acquisition mode; When the second fuzzy factor exceeds the high dynamic range membership threshold and the first fuzzy factor is lower than the motion sensitivity critical value, select the high dynamic range acquisition mode; For the conflict scenario that simultaneously satisfies the strong motion membership condition and the high dynamic range membership condition, calculate the mixing weight through linear interpolation of the motion intensity parameter and the dynamic range parameter; Perform weighted selection on the motion-priority acquisition mode and the high dynamic range acquisition mode according to the mixing weight.

[0008] Optionally, in the third implementation manner of the first aspect of the present application, the step of adjusting the exposure parameter of the high-frame-rate image sensor according to the selected acquisition mode includes: When the motion-priority acquisition mode is selected, generate an exposure time decay coefficient based on the time derivative of the motion intensity parameter; Perform integral operation on the decay coefficient through a time-based function to generate an exposure time sequence; When the high dynamic range acquisition mode is selected, determine the adjacent exposure parameter difference according to the gradient change rate of the dynamic range parameter, and generate a multi-exposure parameter combination based on the adjacent exposure parameter difference; When the hybrid acquisition mode is selected, the exposure time series and the multi-exposure parameter combination are dynamically interpolated by the hybrid weight to generate hybrid exposure parameters.

[0009] Optionally, in the fourth implementation manner of the first aspect of the present application, the step of fusing the optical flow velocity and the luminance of the image stream into a weighting factor according to the exposure parameter and injecting the weighting factor into the image space to generate a scene partition mask includes: Performing an optical flow field calculation on consecutive frames of the image stream according to the exposure parameter to generate a set of optical flow velocity vectors for each pixel point; Generating an initial weighting factor matrix by nonlinearly superimposing the set of optical flow velocity vectors and the luminance gradient of the corresponding pixels; Performing an anisotropic diffusion filtering process on the initial weighting factor matrix to generate a smooth weight distribution map with spatial continuity constraints; Performing binary region segmentation on the smooth weight distribution map according to a preset dynamic threshold to generate a scene partition mask.

[0010] Optionally, in the fifth implementation manner of the first aspect of the present application, the step of fusing the exposure images into a target acquisition image by aligning the scene partition mask with exposure images of different exposure parameters includes: Performing multi-scale decomposition on the exposure images of different exposure parameters according to the edge sharpness parameter of the scene partition mask to determine a set of dual-channel images; wherein the set of dual-channel images includes a high-frequency detail layer and a low-frequency base color layer; Performing motion vector tracking on the high-frequency detail layer through the motion-sensitive region identification of the scene partition mask to generate a spatio-temporal domain fusion coefficient matrix corresponding to the exposure parameter sequence; Performing gradient domain fusion operation on the low-frequency base color layer based on the spatio-temporal domain fusion coefficient matrix to generate a synthetic base color layer that meets the luminance consistency constraint; Performing spatial domain compensation on the chrominance channel of the synthetic base color layer according to the dynamic region ratio of the scene partition mask to generate a target acquisition image that matches the color temperature distribution of the exposure image.

[0011] Optionally, in the sixth implementation manner of the first aspect of the present application, the method further includes: Determining an image quality evaluation parameter of the target acquisition image according to the gradient histogram distribution and the noise spectral density of the target acquisition image; Generating an attenuation compensation coefficient of the motion intensity parameter and a gain correction amount of the dynamic range parameter by performing a time-domain difference operation on the image quality evaluation parameter and the environmental state parameter; Nonlinearly remapping the hybrid weight according to the attenuation compensation coefficient; Adjust the brightness growth coefficient of the multi-exposure parameter combination according to the gain correction amount; Iteratively update the acquisition mode switching threshold of the high-frame-rate image sensor through the remapped mixing weight and the adjusted brightness growth coefficient.

[0012] The second aspect of this application provides an image acquisition control device based on machine vision. The image acquisition control device based on machine vision includes: A determination module, configured to continuously acquire an image stream through a high-frame-rate image sensor and determine corresponding environmental state parameters; A selection module, configured to select the acquisition mode of the high-frame-rate image sensor according to the environmental state parameters; An adjustment module, configured to adjust the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; A generation module, configured to fuse the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and inject the weighting factor into the image space to generate a scene partition mask; A fusion module, configured to fuse the exposure images into a target acquisition image by aligning the scene partition mask with the exposure images of different exposure parameters.

[0013] The third aspect of the embodiments of this application provides an electronic device, including a memory and a processor. The processor is configured to execute a computer program stored on the memory. When the processor executes the computer program, it implements the steps in the image acquisition control method based on machine vision provided in the first aspect of the embodiments of this application.

[0014] The fourth aspect of the embodiments of this application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps in the image acquisition control method based on machine vision provided in the first aspect of the embodiments of this application.

[0015] In summary, according to a method, apparatus, and device for image acquisition control based on machine vision provided by the solution of the present application, an image stream is continuously acquired by a high-frame-rate image sensor and corresponding environmental state parameters are determined; a high-frame-rate image sensor acquisition mode is selected according to the environmental state parameters; exposure parameters of the high-frame-rate image sensor are adjusted according to the selected acquisition mode; an optical flow velocity and brightness of the image stream are fused into a weighting factor according to the exposure parameters, and the weighting factor is injected into an image space to generate a scene partition mask; and exposure images with different exposure parameters are aligned by the scene partition mask, and the exposure images are fused into a target acquisition image. By implementing the solution of the present application, environmental state parameters are generated in real time and corresponding acquisition modes are selected, exposure parameters are optimized, optical flow and brightness data are fused to generate a weighting factor, a scene partition mask is created, multi-exposure images are aligned by the scene partition mask, and the best-exposed parts of each region are precisely fused to obtain a high-quality target image, thereby improving image acquisition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic flowchart of a method for image acquisition control based on machine vision provided by an embodiment of the present application; Figure 2 is a schematic diagram of program modules of an apparatus for image acquisition control based on machine vision provided by an embodiment of the present application; Figure 3 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the objectives, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0018] In order to solve the problem that high-quality panoramic images cannot be obtained with simply set exposure parameters and environmental parameters in related technologies, an embodiment of the present application provides a method for image acquisition control based on machine vision, as Figure 1 shown in the schematic flowchart of the method for image acquisition control based on machine vision provided by this embodiment. The method for image acquisition control based on machine vision includes the following steps: Step 110: Continuously acquire an image stream by a high-frame-rate image sensor and determine corresponding environmental state parameters.

[0019] Specifically, a sequence of images is continuously captured by a high - frame - rate image sensor to achieve real - time monitoring of the target scene. The high - speed acquisition characteristic is utilized to ensure the continuity and integrity of the image data. Meanwhile, an optical flow algorithm is adopted to calculate the pixel displacement between consecutive frames, extract motion information, and perform histogram analysis on the luminance component to obtain the illumination change characteristics. Furthermore, the motion intensity and dynamic range information are fused to determine the environmental state parameters. These parameters can comprehensively reflect the motion blur degree and the luminance distribution in the high - dynamic - range scene, providing a scientific basis for subsequent acquisition mode selection and exposure parameter adjustment, thereby effectively solving the challenges brought by high - speed motion and strong light to image acquisition.

[0020] In an alternative implementation manner of this embodiment, the steps of continuously collecting an image stream by a high - frame - rate image sensor and determining the corresponding environmental state parameters include: continuously collecting an image stream by a high - frame - rate image sensor; generating a motion intensity parameter by normalizing the pixel displacement field of the image stream; extracting the luminance component of the image stream, determining the luminance extreme values by bimodal histogram fitting, and performing logarithmic quantization based on the ratio of the luminance extreme values to generate a dynamic range parameter; and fusing the motion intensity parameter and the dynamic range parameter to generate an environmental state parameter.

[0021] Specifically, high-frame-rate image sensors meticulously capture every moment in an image as a sequence of frames. Key to this is the sensor's rapid response to photoelectric conversion, enabling it to acquire continuous image information at rates of hundreds or even thousands of frames per second, thereby capturing minute movements and fleeting changes in objects even in high-speed motion. Thanks to precise synchronization of the internal clock and electronic shutter, the sensor ensures extremely precise time intervals between frames. This temporal resolution enables subsequent motion analysis to be based on a reliable and uniform time base. After obtaining a continuous image stream, the displacement of each pixel position between adjacent frames is first measured to generate a pixel displacement field covering the entire image. Each displacement vector in this displacement field consists of horizontal and vertical components; its length represents the magnitude of the pixel's motion, while its direction reflects the trajectory of the movement. Taking the absolute values of all displacement lengths and summing them together creates a representation of the motion distribution across the entire image. To eliminate scale differences introduced by different scenes and camera shake, the pixel displacement field is normalized. Normalization involves mapping the original displacement values to a fixed range. This is typically done by dividing the displacement of each pixel by the maximum displacement of the entire image, so that the resulting motion intensity parameter falls between 0 and 1, where 0 represents a stationary area and 1 represents the maximum detected motion. Simultaneously, luminance information within the image stream is extracted and quantized to characterize lighting conditions. The luminance component typically refers to the sequence of grayscale values obtained after converting an image from color space to grayscale space. This value represents the pixel's response to visible light intensity. Plotting a histogram of the luminance component of an entire image frame often reveals two distinct peaks: one corresponding to dark regions and the other to bright regions. Fitting this bimodal histogram accurately determines the two extreme brightness values, or the centers of the two peaks in the histogram. The dark extreme value reflects the typical brightness of shadows or low-light areas, while the bright extreme value represents the typical brightness of partially overexposed or brightly lit areas. The ratio of the bright extreme value to the dark extreme value is calculated and its logarithm is taken to facilitate logarithmic quantification of the dynamic range, resulting in a numerical dynamic range parameter. Logarithmic quantization helps convert the multiplicative relationship in brightness and contrast ranges into an additive one, making numerical processing more stable. For example, in outdoor photography, when direct sunlight and shadows coexist, the highlight extreme may be 200, while the shadow extreme is 20. The ratio is 10, which after logarithmization is approximately 1, thus obtaining the dynamic range parameter of the scene.

[0022] When generating environmental state parameters, first convert the motion intensity and dynamic range into comparable metrics respectively to ensure their comparability within the same numerical scale. Then, according to the relative importance of motion and lighting conditions in the current frame, assign reasonable weights to each metric so that the motion metric dominates in high-speed motion scenarios, while the dynamic range metric is more influential in scenarios with prominent lighting contrast. Next, by fusing these two weighted values, obtain a single parameter that comprehensively reflects the scene state. This parameter can not only indicate the risk of motion blur in the frame but also suggest the need to retain details in the bright and dark parts. The fusion process handles the weight assignment in a smooth transition manner to avoid sudden switches under extreme conditions, ensuring the continuity and stability of the output environmental parameters in various scenarios. Finally, this parameter provides an intuitive and reliable basis for the subsequent decisions on exposure strategies and image fusion schemes, enabling the image acquisition system to adaptively balance clarity and detail retention in a complex environment of motion and lighting interaction, thus achieving consistent high-quality imaging effects.

[0023] Step 120: Select the high-frame-rate image sensor acquisition mode according to the environmental state parameter.

[0024] Specifically, intelligently select the image acquisition mode according to the environmental state parameter. By calculating the motion parameter and the dynamic range parameter respectively and processing them with Gaussian and S-shaped membership functions, generate fuzzy factors reflecting motion blur and brightness change, and use the fuzzy rule base to achieve cross-validation. Automatically enable the motion-priority acquisition mode in scenarios with large motion amplitude and small brightness change, while select the high-dynamic-range acquisition mode when the brightness difference is significant. For complex scenarios with both obvious motion and large dynamic range, calculate the mixed weight by linear interpolation to achieve a balanced switch between the two modes, thereby ensuring that image acquisition can take into account both motion clarity and lighting details.

[0025] In an optional implementation manner of this embodiment, the step of selecting the high-frame-rate image sensor acquisition mode according to the environmental state parameter includes: performing Gaussian membership calculation on the motion intensity parameter to generate a first fuzzy factor, and at the same time performing S-shaped membership calculation on the dynamic range parameter to generate a second fuzzy factor; performing cross-validation on the first fuzzy factor and the second fuzzy factor through the fuzzy rule base; when the first fuzzy factor reaches the strong motion membership interval and the second fuzzy factor belongs to the low dynamic range membership interval, select the motion-priority acquisition mode; when the second fuzzy factor exceeds the high dynamic range membership threshold and the first fuzzy factor is lower than the motion sensitivity critical value, select the high-dynamic-range acquisition mode; for conflict scenarios that simultaneously meet the strong motion membership condition and the high dynamic range membership condition, calculate the mixed weight by linear interpolation of the motion intensity parameter and the dynamic range parameter; perform weighted selection on the motion-priority acquisition mode and the high-dynamic-range acquisition mode according to the mixed weight.

[0026] Specifically, in this embodiment, by calculating the exercise intensity parameter using a Gaussian function, a first fuzzy factor is generated after numerical mapping. The principle lies in using the probability density distribution characteristics of the Gaussian function to perform smooth mapping on continuously collected exercise data, assigning a higher membership degree when the exercise data approaches the expected central value and rapidly decaying when it is far from the center, thereby constructing a fuzzy quantity reflecting the intensity of exercise numerically. Here, the Gaussian function is commonly used in physics to describe the distribution of random variables and precisely characterizes exercise characteristics by adaptively adjusting the parameter mean and standard deviation. At the same time, for the dynamic range parameter, an S-shaped membership function is used. The core lies in the slow start and rapid rise characteristics of the S-shaped curve, converting the brightness extreme values obtained by fitting the bimodal histogram into a fuzzy factor with a lower value in the low dynamic range region and rapidly approaching the upper limit in the high dynamic range region. The S-shaped function is commonly used in system control theory for smooth transition and can effectively express the change in brightness contrast in the image. Subsequently, the fuzzy rule base is used to cross-validate the two generated fuzzy factors. The cross-validation process is based on predefined fuzzy logic rules. By comparing the membership degree values in terms of both exercise and brightness, a qualitative judgment of the scene state is achieved. For example, in a traffic monitoring scene, when the exercise factor reaches a high value and the brightness factor remains in a lower range, the rule base will determine that the current scene is in a high-speed movement state, thus indicating the adoption of the exercise-priority acquisition mode to ensure clear vehicle movement trajectories. In scenes with extremely uneven indoor or outdoor lighting and little exercise variation, when the brightness factor exceeds the threshold and the exercise factor is lower than the preset critical value, the system automatically selects the high dynamic range acquisition mode to ensure that light and dark details in the image can be fully presented. In complex scenes with both intense exercise and high dynamic range characteristics, a mixed weight is calculated by linearly interpolating the exercise parameter and the dynamic range parameter to form a mixed index comprehensively reflecting the scene state. The calculation process takes into account the uneven influence of the two parameters in different intervals. In addition, a non-linear fusion model based on the integral form of the beta distribution is introduced, which can more accurately capture the uncertain factors in conflict scenes. The formula for the non-linear fusion model in the integral form of the beta distribution can be expressed as: , wherein, is the beta function, which is used to normalize the integration result to ensure that the obtained mixed weight W is between 0 and 1. represents the exercise membership degree calculated from the Gaussian membership degree, and its value reflects the significance of exercise information in the image. And represents the dynamic range membership degree calculated through the S-shaped membership function, and its value characterizes the level of brightness contrast in the image. At the same time, and are shape parameters, and their expressions can be designed as and , where is a constant adjusted according to the actual environment and experimental data, used to control the influence of brightness information on the overall sensitivity. The integral variable t represents the normalized variable within the range of 0 to 1. The main purpose of this formula is to non-linearly map two membership degrees into a mixing weight in the form of the cumulative distribution function of the beta distribution. Through integral operation, it can capture the uncertainty factors of motion information and brightness information in conflict scenarios, and then, based on the contribution degrees of these two aspects in different scenarios, achieve an adaptive weighted selection of the acquisition mode. For example, in highway monitoring, when the vehicle motion in the image is intense and the lighting conditions are relatively uniform, is relatively high while is at a relatively low level. Then, the mixing weight W obtained through the calculation of this formula will tend to the motion-priority mode, thus ensuring clear edges of moving objects. On the contrary, in an environment with obvious differences in indoor and outdoor lighting, if is much higher than the motion membership degree, the W output by the formula will tend to the high-dynamic-range acquisition mode to retain the details of bright and dark areas in the image. The advantage of the entire model lies in using the unique shape-adjustability of the beta distribution, enabling the shape parameters to be dynamically adjusted according to the brightness membership degree, and then more accurately modeling the non-linear relationship between the two factors of motion and brightness. The integral operation in the formula not only realizes the cumulative effect of the probability distribution but also provides a mathematical basis for information fusion in complex scenarios, thereby achieving adaptive and refined control of environmental state parameters, and significantly improving the performance and stability of the image acquisition system in motion and high-dynamic-range scenarios in practical applications.

[0027] Step 130: Adjust the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode.

[0028] Specifically, adjust the exposure parameters dynamically according to the selected acquisition mode. In the motion-priority mode, calculate the attenuation coefficient of the exposure time through the time derivative of the motion intensity parameter, and use the time basis function for integral operation to generate a continuous exposure time sequence to reduce motion blur. In the high-dynamic-range mode, determine the difference between adjacent exposure parameters according to the gradient change rate of the dynamic range parameter, and construct a multi-exposure parameter combination to capture the detail information of bright and dark areas in the scene. In the hybrid mode, achieve a dynamic balance between the two requirements of motion and lighting through linear interpolation, thereby realizing real-time control of the exposure parameters and ensuring the coordination of details and brightness of the acquired image when the motion and lighting conditions change.

[0029] In an alternative implementation of this embodiment, the steps of adjusting the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode include: when the motion-priority acquisition mode is selected, generating an exposure-time attenuation coefficient based on the time derivative of the motion-intensity parameter; performing an integration operation on the attenuation coefficient through a time-based function to generate an exposure-time sequence; when the high-dynamic-range acquisition mode is selected, determining the difference between adjacent exposure parameters according to the gradient change rate of the dynamic-range parameter, and generating a multi-exposure parameter combination based on the difference between adjacent exposure parameters; when the hybrid acquisition mode is selected, dynamically interpolating the exposure-time sequence and the multi-exposure parameter combination through a hybrid weight to generate hybrid exposure parameters.

[0030] Specifically, based on the previously obtained exercise intensity parameter and dynamic range parameter, first, both are mapped to fuzzy factors between 0 and 1 through a Gaussian membership function and an S-shaped membership function respectively. The Gaussian membership function takes the exercise intensity as the input and calculates the membership degree through a bell-shaped curve characterized by the central value and the distribution width. Higher exercise amounts correspond to higher membership degrees, and when the exercise amount deviates from this center, the membership degree will rapidly decrease, thus reflecting the continuous change in the intensity of movement in the picture. The S-shaped membership function, on the other hand, performs a smooth mapping on the dynamic range parameter. When the brightness difference is small, the function value is low. As the dynamic range increases, the function value approaches 1 in a curve form that starts gently, rises rapidly, and then saturates, so as to characterize the level of the overall brightness contrast of the image. Next, the two obtained fuzzy factors are input into a pre-designed fuzzy rule base for logical evaluation. This rule base contains a series of "if-then" fuzzy logic rules, which give a qualitative judgment by cross-comparing the magnitude relationship between the exercise membership degree and the brightness membership degree. For example, when the exercise membership degree reaches the "strong exercise" interval higher than a certain threshold and the dynamic range membership degree falls within the "low contrast" interval, the system determines that the current scene is mainly characterized by fast movement, so it activates the motion-priority acquisition mode to compress the exposure time to the greatest extent and avoid image blurring. Another example is that when the brightness membership degree exceeds the "high contrast" threshold and the exercise membership degree is in the "low exercise" interval, the logical judgment will set the high-dynamic-range acquisition mode as the first choice to ensure that both bright and dark details can be captured through multiple exposures or a wider exposure range. In the case of a conflict scene where both exercise and dynamic range are in a high-membership state, such as at a sports game site with both fast movement and complex lighting effects, the logical evaluation will trigger a hybrid strategy. At this time, first, linear interpolation is performed on the exercise membership degree and the brightness membership degree in proportion. The interpolated hybrid weight can be understood as a balance coefficient that takes both factors into account. The interpolation process is automatically allocated according to the numerical strength of the two. A higher exercise membership degree will shift the hybrid weight towards motion acquisition, and vice versa towards high-dynamic-range acquisition. Finally, a weighted selection is made between the two modes based on this hybrid weight. If the hybrid weight value is closer to 1, the system will be more inclined to the motion-priority mode; when the value is closer to 0, the system will be more biased towards the high-dynamic-range mode; if it falls within the intermediate interval, the exposure time sequence and multi-exposure parameters will be dynamically combined in proportion, so as to take into account both motion sharpness and brightness details. Through the mapping of fuzzy membership degrees, the cross-validation of the rule base, and the linear interpolation fusion in conflict scenarios, the entire logic can achieve intelligent, continuous, and smooth switching of the acquisition mode under different shooting conditions, ensuring that the imaging is both clear and rich in details.

[0031] Step 140: According to the exposure parameter, fuse the optical flow velocity and the brightness of the image stream into a weighting factor, and inject the weighting factor into the image space to generate a scene partition mask.

[0032] Specifically, according to the adjusted exposure parameters, optical flow calculation is performed on the continuous image stream to obtain the motion velocity vectors of each pixel point, and non-linear superposition operation is carried out in combination with the image brightness gradient to generate an initial weighted factor matrix. Subsequently, anisotropic diffusion filtering is used to smooth the matrix to ensure the preservation of spatial continuity and edge features, and binary segmentation is performed on the smoothed result using a preset dynamic threshold. Finally, the weighted factor is injected into the image space to form a scene partition mask, which can accurately identify the motion-sensitive area and the brightness change area, providing a clear regional segmentation basis for image fusion and improving the accuracy and effect of overall image information processing.

[0033] In an alternative implementation manner of this embodiment, the steps of fusing the optical flow velocity and brightness of the image stream into a weighted factor according to the exposure parameters and injecting the weighted factor into the image space to generate a scene partition mask include: performing optical flow field calculation on the continuous frames of the image stream according to the exposure parameters to generate a set of optical flow velocity vectors of each pixel point; generating an initial weighted factor matrix by non-linearly superposing the set of optical flow velocity vectors and the brightness gradient of the corresponding pixel; performing anisotropic diffusion filtering on the initial weighted factor matrix to generate a smoothed weight distribution map with spatial continuity constraints; and performing binary region segmentation on the smoothed weight distribution map according to the preset dynamic threshold to generate a scene partition mask.

[0034] Specifically, after correcting exposure parameters for a stream of consecutive frames, the system first accurately compares and matches the pixel brightness of adjacent frames to obtain the displacement information of each pixel between the two moments. This system then constructs a set of motion velocity vectors covering the entire image. These vectors contain both horizontal and vertical velocity components, visually depicting the speed and direction of motion at each point in the image. For example, in a highway monitoring scenario, by point-by-point matching of the brightness characteristics of a vehicle body and the background road surface, a velocity map can be generated that reflects the vehicle's trajectory. Regions with higher velocity values correspond to locations where the vehicle is moving faster, while speed values close to zero indicate relative stillness. Next, at each pixel in the velocity map, brightness gradient information is extracted. This information is obtained by measuring the brightness variation within a certain neighborhood around the pixel, representing the strength of edges, textures, or subtle structures. The magnitude of the velocity vector is then nonlinearly superimposed with the corresponding brightness gradient: a higher initial weight is assigned to a pixel that exhibits both high velocity and strong brightness variation; conversely, a lower weight is assigned to a pixel that exhibits neither significant motion nor structural variation. This mapping ensures that regions with significant motion and rich structure are highlighted in the weighting factor matrix, while flat or static regions are relatively deemphasized. This results in an initial grayscale weighting factor matrix that is used to distinguish different areas of focus. After obtaining this initial matrix, it is spatially smoothed to remove noise and maintain the clarity of important edges. This smoothing process is based on weight differences within pixel neighborhoods. Stronger weight fusion is applied to flat regions to reduce random fluctuations, while less fusion is applied to edges or regions with abrupt structural changes to preserve detail. After multiple iterations, a smoothed weight distribution map with spatial continuity constraints is obtained. This effectively removes isolated high-frequency noise while preserving key texture contours, providing a more reliable basis for region segmentation. Subsequently, a threshold is dynamically determined based on the overall statistical characteristics of the smoothed weight distribution map for binary segmentation. This threshold can refer to the mean, variance, or historical weight distribution of the weight values in the current image. Regions with weights exceeding this threshold are marked as regions of interest; otherwise, they are classified as background. Dynamic threshold adjustment ensures adaptability under varying lighting conditions and motion intensities, enabling accurate capture of motion-significant areas in both low-light scenes at night and bright light during the day. After binarization, the resulting black-and-white mask represents the scene partitioning mask. White areas cover locations with both high motion and structural detail, while black areas correspond to relatively still or textured background areas.

[0035] For example, in a shooting scene of a sports competition, the body parts of athletes often have both rapid displacement and obvious edges of clothing texture at the same time. Through the above process, the initial matrix assigns a high weight to this area; after smoothing, the motion boundary remains clear while the background gradually becomes uniform; the mask generated after binarization can clearly separate the athletes from the stadium background, enabling the subsequent multi-exposure fusion or detail enhancement steps to adopt differential lighting and sharpness compensation strategies for the athletes and the background. Thus, even in the case of the intersection of high-speed motion and complex lighting, a high-quality, low-blur synthetic image can still be obtained.

[0036] Step 150: Align the exposure images with different exposure parameters through the scene partition mask, and fuse the exposure images into a target acquisition image.

[0037] Specifically, by using the scene partition mask to align the exposure images collected under different exposure parameters, the multi-scale image decomposition technology is used to decompose the image into a high-frequency detail layer and a low-frequency base color layer. In the high-frequency layer, dynamic details are extracted through motion vector tracking, and in the low-frequency layer, the gradient domain fusion technology is used to achieve brightness consistency. Then, combined with spatial domain compensation, the chrominance channel is adjusted to ensure that the finally fused image maintains a good match with the original image in terms of color temperature and brightness, so as to obtain a target acquisition image with both high detail performance and balanced lighting, meeting the requirements of high-quality image acquisition and subsequent processing.

[0038] In an optional implementation manner of this embodiment, the step of aligning the exposure images with different exposure parameters through the scene partition mask and fusing the exposure images into a target acquisition image includes: performing multi-scale decomposition on the exposure images with different exposure parameters according to the edge sharpness parameter of the scene partition mask to determine a dual-channel image set; where the dual-channel image set includes a high-frequency detail layer and a low-frequency base color layer; performing motion vector tracking on the high-frequency detail layer through the motion-sensitive area identifier of the scene partition mask to generate a spatio-temporal domain fusion coefficient matrix corresponding to the exposure parameter sequence; performing gradient domain fusion operation on the low-frequency base color layer based on the spatio-temporal domain fusion coefficient matrix to generate a synthetic base color layer that meets the brightness consistency constraint; and performing spatial domain compensation on the chrominance channel of the synthetic base color layer according to the dynamic area ratio of the scene partition mask to generate a target acquisition image that matches the color temperature distribution of the exposure image.

[0039] Specifically, the mask edge sharpness parameter is used to measure the clarity of the boundaries of the scene partition mask, reflecting the sharpness of the transition between different regions. In image processing, the scene partition mask is used to distinguish different scene regions, such as static background regions, moving target regions, shadow regions, or high dynamic range regions, etc. The edge sharpness parameter can help determine whether the boundaries between these regions are clear or there are blurred transitions. In this embodiment, after the scene partition mask is generated, in order to achieve effective fusion of images under different exposure parameters, it is necessary to first perform multi-scale decomposition on multiple exposure images according to the mask edge sharpness parameter, so as to construct a two-channel image set, which includes a high-frequency detail layer and a low-frequency base color layer. The high-frequency detail layer retains the edge and texture information in the image, while the low-frequency base color layer mainly carries the global illumination distribution and color information. Specifically, methods such as Laplacian pyramid or wavelet transform are used to perform multi-scale decomposition on the input image, so that the high-frequency detail layer can more accurately depict the texture changes in the scene, while the low-frequency base color layer provides a stable brightness and color reference. For example, in the process of HDR (High Dynamic Range) image processing, the details of the shadow region are often hidden in the low-exposure image, while the true color of the highlight region is more easily extracted from the high-exposure image. Therefore, high-low frequency decomposition can effectively distinguish the scopes of the two. Based on the motion-sensitive regions identified by the scene partition mask, motion vector tracking is applied to the high-frequency detail layer to ensure the consistency of multiple exposure images in the spatio-temporal dimension. Motion vector tracking can adopt an optical flow-based matching method to calculate the pixel displacement field between different exposure images, and use dynamic time warping (DTW) or adaptive weighted matching methods to optimize the motion compensation error, so that the exposure sequence can maintain a relatively accurate alignment relationship in the motion region. Finally, through the obtained motion vector information, a spatio-temporal domain fusion coefficient matrix matching the exposure parameter sequence is constructed. After the spatio-temporal domain fusion coefficient matrix is constructed, it is necessary to perform gradient domain fusion operation on the low-frequency base color layer to ensure brightness consistency. The core idea of gradient domain fusion is to use the Poisson equation to constrain the gradient field of the synthesized image, so that it satisfies smooth brightness transition in the local region, while still retaining the color characteristics of the original exposure image globally. After the base color layer is fused, it is necessary to further consider the dynamic region ratio of the scene partition mask and perform spatial domain compensation on the chrominance components of the synthesized base color layer (such as the a and b channels in the Lab color space) to ensure the color balance of the final target image and the color temperature matching of the exposure image. In the compensation process, according to the distribution of dynamic and static regions, the dynamic regions will be given a stronger color following weight to correct the color deviation caused by motion compensation and multi-exposure fusion, while the static regions focus more on retaining the color accuracy of the original exposure image. Through regional adaptive interpolation, it is possible to eliminate the color temperature drift that may occur due to the fusion process while maintaining color saturation and contrast, and finally obtain a target acquisition image that is both sharp and clear in details and balanced and natural in terms of illumination and color.For example, when shooting a stage performance, the rapidly moving actors may cause a color temperature shift after multi-exposure fusion. By this method, the colors in the dynamic area can be made consistent with the surrounding environment, making the final output image more natural and harmonious in terms of hue.

[0040] In an optional implementation manner of this embodiment, according to the gradient histogram distribution and noise spectral density of the target acquisition image, an image quality evaluation parameter of the target acquisition image is determined; by performing a time-domain difference operation on the image quality evaluation parameter and the environmental state parameter, an attenuation compensation coefficient of the motion intensity parameter and a gain correction amount of the dynamic range parameter are generated; the mixing weight is non-linearly remapped according to the attenuation compensation coefficient; the brightness growth coefficient of the multi-exposure parameter combination is adjusted according to the gain correction amount; the acquisition mode switching threshold of the high-frame-rate image sensor is iteratively updated through the remapped mixing weight and the adjusted brightness growth coefficient.

[0041] Specifically, after obtaining the target acquisition image, it is necessary to analyze its image quality to optimize subsequent acquisition parameter adjustment. The gradient histogram distribution is used to measure the edge sharpness information in the image, while the noise spectral density is used to evaluate the random noise level in the image. The gradient histogram distribution can reflect the change intensity of different brightness regions in the image. For example, in a dynamic scene, if the edge sharpness decreases, it may mean an increase in motion blur; while the noise spectral density can identify signal interference under different exposure conditions through frequency domain analysis. For example, in low-light environments, the noise level of the image is usually high. By combining these two features, an image quality evaluation parameter can be calculated, which is used to measure the overall quality of the target image in terms of sharpness, noise level, etc. After obtaining the image quality evaluation parameter, it is necessary to conduct a temporal comparative analysis with the environmental state parameters to determine the attenuation compensation coefficient of the motion intensity and the gain correction amount of the dynamic range. The temporal comparative analysis can reveal the change trend of the image quality. For example, in a fast-moving scene, if the image quality deteriorates, it means an increase in motion blur. At this time, it is necessary to adjust the motion compensation parameters; while in a scene with drastic brightness changes, if the distribution of bright and dark regions in the image is uneven, it is necessary to correct the dynamic range parameters to ensure exposure stability. For example, in a scene of tracking a high-speed vehicle, if the image quality drops sharply in a short period of time, it indicates that the acquisition system needs to strengthen motion compensation. And in a night surveillance scene, if the noise level increases, it is necessary to enhance the dynamic range to ensure the visibility of dark details. To optimize the weights between different acquisition modes, it is necessary to perform a non-linear adjustment on the hybrid weights based on the attenuation compensation coefficient of the motion intensity. The hybrid weights are used to determine the fusion ratio of the motion priority mode and the high dynamic range mode, and the non-linear adjustment can ensure the smooth transition of the weights under different environmental changes. For example, during the live broadcast of a sports game, when the camera tracks a fast-moving athlete, the system will automatically tend to the motion priority mode to reduce motion blur; while when the athlete stops moving, the system will gradually switch to the high dynamic range mode to enhance the detail expression of the picture. Through this non-linear adjustment, the system can dynamically adapt to changes in different scenes and ensure the optimal final image quality. While adjusting the hybrid weights, it is also necessary to optimize the brightness adjustment coefficient between different exposure images according to the gain correction amount of the dynamic range. The brightness adjustment coefficient is used to control the brightness contrast of each frame image in the multi-exposure acquisition mode to ensure the exposure balance of the final synthesized image. For example, in the scene of the tunnel entrance and exit, the change in lighting conditions may cause overexposure or underexposure problems in the image. At this time, the system needs to optimize the transition of the exposure parameters according to the brightness adjustment coefficient, so that the final output image can retain the details of the bright part and ensure the visibility of the dark part. Finally, the optimized hybrid weights and brightness adjustment coefficient will be used to iteratively update the acquisition mode switching threshold of the high-frame-rate image sensor to improve the self-adaptability of the system.Iterative updates can dynamically adjust the switching logic of the acquisition mode based on historical frame data to avoid sudden mode changes. For example, in urban traffic monitoring, the lighting conditions during the day and at night vary significantly. The system can adjust the activation threshold of the high-dynamic range mode according to the long-term trend to ensure the imaging quality at night. In scenes with intense movement, the switching threshold of the motion-priority mode can be appropriately reduced to ensure the stability of the picture clarity. Through this adaptive iterative optimization, the system can achieve more intelligent exposure control in different scenarios, thereby improving the visual quality of the final image.

[0042] A machine vision-based image acquisition control method provided by the solution of the present application selects a high-frame-rate image sensor acquisition mode according to environmental state parameters; adjusts the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; fuses the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and injects the weighting factor into the image space to generate a scene partition mask; and fuses the exposure images into a target acquisition image by aligning the scene partition mask with the exposure images of different exposure parameters. Through the implementation of the solution of the present application, environmental state parameters are generated in real time and the corresponding acquisition mode is selected, the exposure parameters are optimized, the optical flow and brightness data are fused to generate a weighting factor, a scene partition mask is created, the multi-exposure images are aligned through the scene partition mask, and the best-exposed parts of each region are precisely fused to obtain a high-quality target image, thereby improving the image acquisition efficiency.

[0043] Figure 2 A machine vision-based image acquisition control device provided by an embodiment of the present application can be used to implement the machine vision-based image acquisition control method in the foregoing embodiments. As Figure 2 shown, the machine vision-based image acquisition control device mainly includes: A determination module 10 for continuously acquiring an image stream through a high-frame-rate image sensor and determining the corresponding environmental state parameters; A selection module 20 for selecting a high-frame-rate image sensor acquisition mode according to environmental state parameters; An adjustment module 30 for adjusting the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; A generation module 40 for fusing the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and injecting the weighting factor into the image space to generate a scene partition mask; A fusion module 50 for fusing the exposure images into a target acquisition image by aligning the scene partition mask with the exposure images of different exposure parameters.

[0044] In an optional implementation manner of this embodiment, the determination module is specifically configured to: continuously collect an image stream through a high-frame-rate image sensor; generate a motion intensity parameter by normalizing the pixel displacement field of the image stream; extract the luminance component of the image stream, determine the luminance extreme value by fitting a bimodal histogram, perform logarithmic quantization based on the ratio of the luminance extreme values, and generate a dynamic range parameter; fuse the motion intensity parameter and the dynamic range parameter to generate an environmental state parameter.

[0045] In an optional implementation manner of this embodiment, the selection module is specifically configured to: perform Gaussian membership calculation on the motion intensity parameter to generate a first fuzzy factor, and at the same time perform S-type membership calculation on the dynamic range parameter to generate a second fuzzy factor; perform cross-validation on the first fuzzy factor and the second fuzzy factor through a fuzzy rule base; when the first fuzzy factor reaches the strong motion membership interval and the second fuzzy factor belongs to the low dynamic range membership interval, select the motion-priority acquisition mode; when the second fuzzy factor exceeds the high dynamic range membership threshold and the first fuzzy factor is lower than the motion sensitivity critical value, select the high dynamic range acquisition mode; for the conflict scenario that simultaneously satisfies the strong motion membership condition and the high dynamic range membership condition, calculate the mixing weight through linear interpolation of the motion intensity parameter and the dynamic range parameter; perform weighted selection on the motion-priority acquisition mode and the high dynamic range acquisition mode according to the mixing weight.

[0046] In an optional implementation manner of this embodiment, the adjustment module is specifically configured to: when the motion-priority acquisition mode is selected, generate an exposure time attenuation coefficient based on the time derivative of the motion intensity parameter; perform integral operation on the attenuation coefficient through a time basis function to generate an exposure time sequence; when the high dynamic range acquisition mode is selected, determine the adjacent exposure parameter difference according to the gradient change rate of the dynamic range parameter, and generate a multi-exposure parameter combination based on the adjacent exposure parameter difference; when the hybrid acquisition mode is selected, perform dynamic interpolation on the exposure time sequence and the multi-exposure parameter combination through the mixing weight to generate a hybrid exposure parameter.

[0047] In an optional implementation manner of this embodiment, the generation module is specifically configured to: perform an optical flow field calculation on consecutive frames of the image stream according to the exposure parameter to generate a set of optical flow velocity vectors for each pixel point; generate an initial weighted factor matrix by non-linearly superimposing the set of optical flow velocity vectors and the luminance gradient of the corresponding pixel; perform anisotropic diffusion filtering on the initial weighted factor matrix to generate a smoothed weight distribution map with spatial continuity constraints; perform binary region segmentation on the smoothed weight distribution map according to a preset dynamic threshold to generate a scene partition mask.

[0048] In an alternative implementation of this embodiment, the fusion module is specifically configured to: perform multi-scale decomposition on exposure images with different exposure parameters according to the edge sharpness parameter of the scene partition mask to determine a dual-channel image set; wherein, the dual-channel image set includes a high-frequency detail layer and a low-frequency base color layer; perform motion vector tracking on the high-frequency detail layer through the motion-sensitive area identifier of the scene partition mask to generate a spatio-temporal domain fusion coefficient matrix corresponding to the exposure parameter sequence; perform gradient domain fusion operation on the low-frequency base color layer based on the spatio-temporal domain fusion coefficient matrix to generate a synthetic base color layer that meets the brightness consistency constraint; perform spatial domain compensation on the chrominance channel of the synthetic base color layer according to the dynamic area ratio of the scene partition mask to generate a target acquisition image that matches the color temperature distribution of the exposure image.

[0049] In an alternative implementation of this embodiment, the image acquisition control device further includes: an optimization module. The optimization module is configured to: determine an image quality evaluation parameter of the target acquisition image according to the gradient histogram distribution and noise spectral density of the target acquisition image; generate an attenuation compensation coefficient for the motion intensity parameter and a gain correction amount for the dynamic range parameter by performing a time-domain difference operation on the image quality evaluation parameter and the environmental state parameter; perform non-linear remapping on the mixing weight according to the attenuation compensation coefficient; adjust the brightness growth coefficient of the multi-exposure parameter combination according to the gain correction amount; and iteratively update the acquisition mode switching threshold of the high-frame-rate image sensor through the remapped mixing weight and the adjusted brightness growth coefficient.

[0050] An image acquisition control device based on machine vision provided by the solution of the present application selects a high-frame-rate image sensor acquisition mode according to environmental state parameters; adjusts the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; fuses the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and injects the weighting factor into the image space to generate a scene partition mask; aligns the exposure images with different exposure parameters through the scene partition mask, and fuses the exposure images into a target acquisition image. Through the implementation of the solution of the present application, environmental state parameters are generated in real time and corresponding acquisition modes are selected, exposure parameters are optimized, optical flow and brightness data are fused to generate a weighting factor, a scene partition mask is created, multi-exposure images are aligned through the scene partition mask, and the best-exposed parts of each region are precisely fused to obtain a high-quality target image, thereby improving the image acquisition efficiency.

[0051] According to the solution provided by the present application Figure 3 An electronic device provided in an embodiment of the present application. This electronic device can be used to implement the machine vision-based image acquisition control method in the foregoing embodiments, and mainly includes: A memory 301, a processor 302, and a computer program 303 stored on the memory 301 and executable on the processor 302. The memory 301 and the processor 302 are communicatively connected. When the processor 302 executes the computer program 303, the image acquisition control method based on machine vision in the foregoing embodiments is implemented. Among them, the number of processors can be one or more.

[0052] The memory 301 may be a high-speed random access memory (RAM), or a non-volatile memory, such as a disk memory. The memory 301 is used to store executable program codes, and the processor 302 is coupled to the memory 301.

[0053] Furthermore, an embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium may be disposed in the electronic device in the foregoing embodiments. The computer-readable storage medium may be the Figure 3 memory in the foregoing embodiments.

[0054] A computer program is stored on the computer-readable storage medium. When the program is executed by a processor, the image acquisition control method based on machine vision in the foregoing embodiments is implemented. Furthermore, the computer-readable storage medium may also be various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a RAM, a magnetic disk, or an optical disc that can store program codes.

[0055] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0056] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0057] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An image acquisition control method based on machine vision, characterized in that, Including: Continuously acquiring an image stream through a high-frame-rate image sensor and determining corresponding environmental state parameters; Selecting the acquisition mode of the high-frame-rate image sensor according to the environmental state parameters; Adjusting the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode; Fusing the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and injecting the weighting factor into the image space to generate a scene partition mask; Aligning the scene partition mask with exposure images of different exposure parameters, and fusing the exposure images into a target acquisition image.

2. The image acquisition control method based on machine vision according to claim 1, characterized in that, The step of continuously acquiring an image stream through a high-frame-rate image sensor and determining corresponding environmental state parameters includes: Continuously acquiring an image stream through a high-frame-rate image sensor; Generating the motion intensity parameter by normalizing the pixel displacement field of the image stream; Extracting the luminance component of the image stream, determining the luminance extreme value by bimodal histogram fitting, and performing logarithmic quantization based on the ratio of the luminance extreme values to generate the dynamic range parameter; Fusing the motion intensity parameter and the dynamic range parameter to generate an environmental state parameter.

3. The image acquisition control method based on machine vision according to claim 2, characterized in that, The step of selecting the acquisition mode of the high-frame-rate image sensor according to the environmental state parameters includes: Performing Gaussian membership calculation on the motion intensity parameter to generate a first fuzzy factor, and simultaneously performing S-type membership calculation on the dynamic range parameter to generate a second fuzzy factor; Performing cross-validation on the first fuzzy factor and the second fuzzy factor through a fuzzy rule base; When the first fuzzy factor reaches the strong motion membership interval and the second fuzzy factor belongs to the low dynamic range membership interval, selecting the motion-priority acquisition mode; When the second fuzzy factor exceeds the high dynamic range membership threshold and the first fuzzy factor is lower than the motion sensitivity critical value, selecting the high dynamic range acquisition mode; For a conflict scene that simultaneously satisfies the strong motion membership condition and the high dynamic range membership condition, calculating a mixing weight through linear interpolation of the motion intensity parameter and the dynamic range parameter; Performing weighted selection on the motion-priority acquisition mode and the high dynamic range acquisition mode according to the mixing weight.

4. The image acquisition control method based on machine vision according to claim 3, characterized in that The step of adjusting the exposure parameters of the high-frame-rate image sensor according to the selected acquisition mode includes: When selecting the motion-priority acquisition mode, generating an exposure time decay coefficient based on the time derivative of the motion intensity parameter; Performing integral operation on the decay coefficient through a time base function to generate an exposure time sequence; When selecting the high dynamic range acquisition mode, determining the adjacent exposure parameter difference according to the gradient change rate of the dynamic range parameter, and generating a multi-exposure parameter combination based on the adjacent exposure parameter difference; When selecting the hybrid acquisition mode, performing dynamic interpolation on the exposure time sequence and the multi-exposure parameter combination through the mixing weight to generate hybrid exposure parameters.

5. The image acquisition control method based on machine vision according to claim 1, characterized in that, The step of fusing the optical flow velocity and brightness of the image stream into a weighting factor according to the exposure parameters, and injecting the weighting factor into the image space to generate a scene partition mask includes: Performing optical flow field calculation on consecutive frames of the image stream according to the exposure parameters to generate a set of optical flow velocity vectors for each pixel point; An initial weighted factor matrix is generated by non-linearly superposing the set of optical flow velocity vectors and the luminance gradient of the corresponding pixels; Anisotropic diffusion filtering is performed on the initial weighted factor matrix to generate a smoothed weight distribution map with spatial continuity constraints; Binary region segmentation is performed on the smoothed weight distribution map according to a preset dynamic threshold to generate a scene partition mask; 6. The image acquisition control method based on machine vision according to claim 1, characterized in that, The step of aligning the scene partition mask with the exposure images of different exposure parameters and fusing the exposure images into a target acquisition image includes: Multi-scale decomposition is performed on the exposure images of different exposure parameters according to the edge sharpness parameter of the scene partition mask to determine a dual-channel image set; wherein, the dual-channel image set includes a high-frequency detail layer and a low-frequency base color layer; Motion vector tracking is performed on the high-frequency detail layer through the motion-sensitive region identifier of the scene partition mask to generate a spatio-temporal domain fusion coefficient matrix corresponding to the exposure parameter sequence; Gradient domain fusion operation is performed on the low-frequency base color layer based on the spatio-temporal domain fusion coefficient matrix to generate a synthesized base color layer that conforms to the luminance consistency constraint; According to the dynamic region ratio of the scene partition mask, spatial domain compensation is performed on the chrominance channel of the synthesized base color layer to generate a target acquisition image that matches the color temperature distribution of the exposure image; 7. The image acquisition control method based on machine vision according to claim 4, characterized in that, The method further includes: Determining an image quality evaluation parameter of the target acquisition image according to the gradient histogram distribution and noise spectral density of the target acquisition image; Performing time-domain difference operation on the image quality evaluation parameter and the environmental state parameter to generate an attenuation compensation coefficient of the motion intensity parameter and a gain correction amount of the dynamic range parameter; Non-linearly remapping the mixing weight according to the attenuation compensation coefficient; Adjusting the luminance growth coefficient of the multi-exposure parameter combination according to the gain correction amount; Iteratively updating the acquisition mode switching threshold of the high-frame rate image sensor through the remapped mixing weight and the adjusted luminance growth coefficient; 8. An image acquisition control device based on machine vision, characterized in that, The image acquisition control device based on machine vision includes: A determination module for continuously acquiring an image stream through a high-frame rate image sensor and determining a corresponding environmental state parameter; A selection module for selecting the acquisition mode of the high-frame rate image sensor according to the environmental state parameter; An adjustment module for adjusting the exposure parameter of the high-frame rate image sensor according to the selected acquisition mode; A generation module for fusing the optical flow velocity and luminance of the image stream into a weighted factor according to the exposure parameter and injecting the weighted factor into the image space to generate a scene partition mask; A fusion module for aligning the scene partition mask with the exposure images of different exposure parameters and fusing the exposure images into a target acquisition image; 9. An electronic device, characterized in that, Including a memory and a processor, wherein: The processor is used to execute a computer program stored on the memory; When the processor executes the computer program, the steps in the image acquisition control method based on machine vision according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps in the machine vision-based image acquisition control method according to any one of claims 1 to 7.

Images