An image processing method and device in a long-distance weak light scene and a storage medium

Abstract

The application relates to the field of image processing, in particular to an image processing method and device in a long-distance weak light scene and a storage medium, the method comprising: collecting initial image data based on all pixel units of a large-aperture telescope system at a first preset period, and processing the initial image data based on a preset image processing algorithm to output a first monitoring image; determining a plurality of first interest regions in the first monitoring image, each first interest region corresponding to a group of target pixel units; in the first preset period, performing: predicting residual computing power based on the target pixel units to select a supplementary monitoring region; each supplementary monitoring region corresponding to a group of monitoring pixel units; collecting target image data based on the target pixel units and / or monitoring image data based on the monitoring pixel units, and processing the image data based on the preset image processing algorithm to output a second monitoring image. The imaging quality and response speed of the key region are ensured under limited computing power, and the overall monitoring range is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing, in particular to an image processing method and device in a long-distance low-light scene and a storage medium. BACKGROUND

[0002] The long-distance low-light scene refers to the application requirement of clearly imaging and identifying a target hundreds of meters to several kilometers away under the conditions of night, dusk, dawn or severe weather (such as fog, haze, rain and snow). The long-distance low-light scene widely exists in the fields of border defense monitoring, maritime supervision, forest fire prevention, unmanned aerial vehicle reconnaissance, emergency rescue and astronomical exploration. Due to the long distance of the target and the dim light, the light reaching the sensor is extremely weak, and the imaging generally has problems such as low brightness, poor signal-to-noise ratio and blurred details. At present, the imaging quality in such a scene is usually improved through software algorithms.

[0003] For example, the patent application with the publication number CN112288652A discloses a guided filter Retinex low-illumination image enhancement method, including steps one, converting the original image from RGB space to HSV space; two, one layer wavelet decomposition of the luminance component; three, two layer wavelet decomposition of the luminance component; four, linear enhancement of the high frequency component; five, wavelet reconstruction to get a first approximation image; six, using adaptive median filtering to enhance the high frequency component of one layer decomposition; seven, guided filter Retinex enhances the first approximation image; eight, wavelet reconstruction of the enhanced image; nine, combining the enhanced luminance component with the hue component and the saturation component, and then converting the combined image to RGB space and gamma correction to get the final image.

[0004] For example, the patent application with the publication number CN120198313A discloses a high-efficiency image processing and feature extraction method, which belongs to the field of image processing technology. By introducing adaptive illumination normalization, multi-channel joint denoising and edge-oriented super-resolution reconstruction, etc., the method can effectively deal with complex lighting, noise interference and insufficient resolution, and significantly improve the image quality.

[0005] However, due to the poor quality of the original image in the long-distance low-light scene, the algorithm often needs to undertake a strong denoising and detail reconstruction task, and the computational complexity rises. In actual operation, it usually needs strong computing resources support, and the practicability is low. SUMMARY

[0006] The main purpose of the present application is to provide an image processing method and device in a long-distance low-light scene and a storage medium. In order to solve the above-mentioned technical problems, the present application specifically adopts the following technical solutions:

[0007] The first aspect of the present application provides an image processing method in a long-distance weak light scene, characterized in application to a large-aperture telescope system, the method comprising:

[0008] S101, acquiring initial image data based on all pixel units of the large-aperture telescope system at a first preset period, and processing the initial image data based on a preset image processing algorithm to output a first monitoring image;

[0009] S102, analyzing the brightness change in the initial image data based on a first mutation standard to detect a brightness mutation region, and determining a plurality of first interest regions in the first monitoring image based on the brightness mutation region, each first interest region corresponding to a group of target pixel units;

[0010] Steps S1031-S1032 are performed within the first preset period, comprising:

[0011] S1031, selecting a supplementary monitoring region based on the remaining computing power available based on the target pixel units, wherein each supplementary monitoring region corresponds to a group of monitoring pixel units;

[0012] S1032, acquiring target image data based on the target pixel units of the large-aperture telescope system and / or monitoring image data based on the monitoring pixel units, and processing the target image data and / or the monitoring image data based on a preset image processing algorithm to output a second monitoring image.

[0013] In some embodiments, the large-aperture telescope system comprises a large-aperture telescope of a preset resolution, an AR sheet of a preset waveband, a microlens array, and a CMOS image sensor; the large-aperture telescope comprises a concave primary mirror, a plane secondary mirror, and an eyepiece, a lens group is arranged between the concave primary mirror and the eyepiece to form a relay surface structure to expand the imaging range of the eyepiece focusing; the AR sheet of the preset waveband is arranged in front of the microlens array to filter incident light; the microlens array is used to redirect or focus incident light to a corresponding preset position; and the CMOS image sensor is used to photosensitive all or part of the pixel units to obtain corresponding image data.

[0014] In some embodiments, the preset image processing algorithm comprises a type of processing algorithm, which is used to process the initial image data and the target image data.

[0015] In some embodiments, the preset image processing algorithm further comprises a second type of processing algorithm, the required computing power of the first type of processing algorithm is greater than the required computing power of the second type of processing algorithm, and the S1032 comprises: based on the second type of processing algorithm, performing motion detection according to historical monitoring images and the monitoring image data to determine whether there is an anomaly in the supplementary monitoring area; if yes, updating the second monitoring image based on processing of the monitoring image data of the corresponding supplementary monitoring area by the first type of processing algorithm; or performing steps S101 to S102.

[0016] In some embodiments, the first type of processing algorithm comprises a wavelet filtering denoising algorithm, a multi-scale Retinex image enhancement algorithm, and an edge enhancement algorithm executed in sequence; and the second type of processing algorithm comprises an inter-frame difference algorithm and / or a background modeling algorithm.

[0017] In some embodiments, the method comprises: obtaining a plurality of preset second interest areas, the second interest areas being high-risk areas where there are key monitoring objects or potential luminance mutation objects; and polling at least one second interest area from the plurality of second interest areas as a supplementary monitoring area at a second preset period; wherein the second preset period is less than the first preset period.

[0018] In some embodiments, the method comprises: pre-calibrating a plurality of second interest areas; or identifying at least one key monitoring object from the first monitoring image based on a preset target recognition algorithm, and taking an image area where the key monitoring object is located as a second interest area; or analyzing luminance changes in the initial image data based on a second mutation standard, and taking a detected potential luminance mutation area as a second interest area; wherein an index requirement of the second mutation standard is lower than an index requirement of the first mutation standard.

[0019] In some embodiments, the S1031 comprises: analyzing a luminance change rule of the first interest area based on historical monitoring images, identifying a pseudo-interest area in the first interest area based on the luminance change rule; polling an effective pseudo-interest area from the plurality of pseudo-interest areas at a third preset period, and updating the first interest area; wherein the third preset period is less than the first preset period.

[0020] A second aspect of the present application provides a computer device, the device comprising:

[0021] a memory for storing a computer program;

[0022] a processor for executing the computer program and implementing the steps of the image processing method in a long-distance weak light scene as provided in any of the embodiments of the present application when executing the computer program.

[0023] The third aspect of the present application is also to correspondingly provide a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program is executed by a processor to make the processor execute the steps of the image processing method in a long-distance weak light scene provided by any embodiment of the present application.

[0024] Advantages:

[0025] The embodiments of the present application provide an image processing method, device and storage medium in a long-distance weak light scene, and specifically provide an algorithm power elastic allocation mechanism in image processing. Under limited algorithm power, the imaging quality and response speed of a key area are guaranteed, the perception ability of a monitoring system to a global anomaly and the overall monitoring breadth are improved, and the mechanism is especially suitable for scenes with high algorithm power demand under long-distance weak light conditions, such as regional security monitoring and astronomical observation.

[0026] Specifically, a complete and fine first type of processing algorithm is executed on a full field of view image in a first preset period, the initial image quality is ensured, the accuracy of first interest area identification in each period is improved, and missing detection or misjudgment of a key target caused by image quality is avoided. Further, the interest area is continuously processed with high precision, the algorithm power overhead caused by repeated processing of the full image is reduced, the remaining available algorithm power of the system is dynamically evaluated, a supplementary monitoring area is flexibly selected according to the evaluation result, and fine scheduling and maximum utilization of algorithm resources are realized.

[0027] For the supplementary monitoring area, a lightweight second type of processing algorithm is used to perform low-overhead anomaly preliminary screening, and the first type of processing algorithm is triggered for high-precision processing or a new round of full field of view acquisition and analysis is started only when an anomaly is found, so that fast response to an abnormal event is realized under limited algorithm power, and the global perception ability and sensitivity of the monitoring system are improved.

[0028] Meanwhile, a multi-level polling mechanism is provided to realize dynamic updating and optimization of the monitoring area, so that limited algorithm power is dynamically allocated to image areas with higher value, and the overall monitoring breadth is improved. Firstly, a second interest area is preset or dynamically generated, and the second interest area is polled and monitored in a second preset period, so that the coverage degree of a high-risk area is strengthened. Secondly, a pseudo interest area generated due to suspected environmental interference is periodically polled and verified, so that algorithm power is not wasted on non-threat targets for a long time. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. In all the drawings, similar elements or parts are generally identified by similar reference signs. In the drawings, each element or part is not necessarily drawn according to the actual proportion. Obviously, the drawings described below are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0030] Figure 1 is a schematic flow chart of an image processing method in a long-distance weak light scene provided by an embodiment of the present application;

[0031] Figure 2 is a schematic diagram of a local initial image of a first region of interest provided by an embodiment of the present application;

[0032] Figure 3 is a schematic diagram of a local monitoring image of a first region of interest provided by an embodiment of the present application;

[0033] Figure 4 is a running flow chart of a monitoring imaging system provided by an embodiment of the present application;

[0034] Figure 5 is a structural schematic block diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0036] The flow chart shown in the drawings is only an example description, not necessarily including all the contents and operations / steps, and not necessarily executed in the described order. For example, some operations / steps can also be decomposed, combined or partially merged, so that the actual execution order can be changed according to the actual situation.

[0037] In this paper, the suffix such as "module", "component" or "unit" used to represent elements is only for the convenience of the description of the present application, and has no specific meaning. Therefore, "module", "component" or "unit" can be mixedly used.

[0038] In this document, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0039] In this document, unless otherwise explicitly specified and limited, the terms "mount", "provided with", "connected" and the like should be broadly understood, for example, "connected" can be fixedly connected, can be detachably connected, or integrally connected; can be mechanically connected, can be directly connected, can be indirectly connected through an intermediate medium, can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0040] In this document, "and / or" includes any and all combinations of one or more listed related items.

[0041] In this document, "a plurality of" means two or more, that is, including two, three, four, five, etc.

[0042] It should be noted that in this document, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or device. Without more limitations, the element defined by the sentence "including a…" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0043] In this document, a large-aperture telescope system refers to an optical observation system integrated by a high-resolution large-aperture telescope and a high-performance imaging camera, which can realize clear imaging and identification of weak light targets hundreds of meters to several kilometers away, and is widely used in astronomical observation, security monitoring, maritime supervision and long-distance reconnaissance scenes.

[0044] The embodiment of the application provides a large-aperture telescopic system, which comprises a large-aperture telescope with a preset resolution, an AR sheet with a preset waveband, a microlens array and a CMOS image sensor; the large-aperture telescope comprises a concave primary mirror, a plane secondary mirror and an eyepiece, a lens group is arranged between the concave primary mirror and the eyepiece to form a relay plane structure, so that the imaging range is expanded by focusing the eyepiece; the AR sheet with the preset waveband is arranged in front of the microlens array and is used for filtering incident light; the microlens array is used for redirecting or focusing the incident light to a corresponding preset position; and the CMOS image sensor is used for photosensing all or part of pixel units to obtain corresponding image data.

[0045] The relay plane structure refers to a plane where a relay mirror is located. The relay mirror is a kind of image element used for transferring a light beam or in an optical system, which can change the propagation direction of the light beam, adjust the focusing position of the light beam, etc., so as to meet different observation requirements of the telescope and improve the imaging quality.

[0046] The AR sheet is an anti-reflective coating (AR), which is a kind of surface optical coating layer and can increase the transmittance by reducing the reflection of light. The preset waveband can be flexibly set according to actual requirements.

[0047] In some embodiments, the large-aperture telescope can be a Newtonian reflector, and an example can be an LX90-ACF astronomical telescope. The working process of the Newtonian reflector is that the light of an object is incident from the opening of a lens barrel, is reflected by a concave primary mirror at the bottom of the lens barrel to converge, the focus point of the convergence is originally located near the opening of the lens barrel, a small plane mirror (i.e., a secondary mirror) is installed in front of the focus point by Newton, the light is reflected by 90 degrees from the side at an angle of 45 degrees, and an eyepiece is installed on the side to observe. The main structure of the Newtonian reflector is the concave primary mirror, which is used for converging the light from a remote celestial body, and then reflecting the light on the plane secondary mirror. The plane secondary mirror is used for folding the reflected light into a focusing seat to focus with the eyepiece, and the magnified image can be observed through the eyepiece. The relay plane structure for imaging is arranged between the primary mirror focal plane and the eyepiece in the system, which is composed of a lens group, so that the function of focusing the eyepiece is realized, the depth of field is increased, and therefore the imaging range is increased.

[0048] It should be understood that the same caliber Newtonian reflector is lower in cost than a refractor, and because the light does not enter the interior of the glass, there is no chromatic aberration, and the true colors and details of the object can be presented more clearly. Since the Newtonian reflector uses a concave mirror as the objective lens, and the concave mirror can be made of metal material, thereby breaking the bottleneck of caliber, it is easy to obtain a larger caliber, so as to collect more light, improve the observation sensitivity and resolution, and is suitable for observing dark and weak light objects, and is highly adapted to long-distance weak light scenes.

[0049] In some embodiments, the microlens array is coupled with an AR window piece customized for a visible light band. The AR window piece can be an anti-reflective coating filter, for example, an anti-reflective coating filter with a diameter of 21 mm and a thickness of 1.1 mm, which increases the transmission of light in the 400 nm-720 nm band and is arranged in front of the microlens array. The core principle is to use the interference effect of light to offset the reflected light of the rest of the surface in other bands, and to increase the transmission of light in a specific band. Through a multi-layer coating scheme, a plurality of medium films with different refractive indices are generated, and by accurately controlling the film thickness and material refractive index, the reflectivity of a specific band is minimized. Since visible light contains a variety of different wavelengths of light, the multi-layer coating can maximize the interference effect of different wavelengths of visible light, thereby reducing reflected light and increasing the transmittance of a specific band. When the light is reflected twice on the film of the AR piece, the twice-reflected light will interfere with the original reflected light. If the thickness of the film is equal to one-quarter of the wavelength of a specific wavelength of light in the anti-reflection film, then the light of that wavelength reflected back on both sides of the film will cancel each other out (the optical path difference is an odd multiple of half the wavelength), according to the law of conservation of energy, the reflected light is reduced, and the transmitted light is increased, thereby increasing the transmittance of the light of that wavelength, making the imaging of the optical system clearer. By designing the thickness and material of the film, the visible light in a specific band satisfies the above interference cancellation condition, and the transmittance of the light in that band is high.

[0050] In some embodiments, the microlens array is composed of a small lens array arranged in a regular grid pattern, each microlens has its own focal point, and they work together to manipulate the behavior of light passing through them, allowing the microlens array to control the distribution, direction, and intensity of light. A single microlens can concentrate light and change the direction of light, and when light passes through the microlens array, each microlens will collect and focus the incident light to a specific point in space, i.e., a preset position. The microlens array is used in a large-aperture telescope system to improve light collection efficiency by focusing incident light onto a photodetector (i.e., a preset position), thereby improving image quality and sensitivity. Coupling the microlens array with an AR window piece customized for a visible light band can overcome the limitations of conventional AR pieces that average all bands, and is particularly suitable for scenes with weak light imaging and high contrast display.

[0051] In some embodiments, the CMOS image sensor is a solid-state imaging device that converts optical signals into electrical signals, integrated with pixel array, readout circuit and control logic on a single chip. Each pixel unit thereof contains a light-sensitive element (such as a photodiode) and at least one transistor for photoelectric conversion and signal readout. In the working mode of partial pixel unit photosensing, only the pixels in a specific area (such as the first area of interest and the supplementary monitoring area) are activated for exposure and reading, and the full array is not involved, thereby improving the frame rate, reducing the data volume and power consumption, and being suitable for application scenarios of dynamic tracking of target areas in long-distance monitoring.

[0052] The embodiments of the present application also provide an image processing method in a long-distance low-light scene, and specifically provide an algorithm elasticity allocation mechanism in image processing, which not only guarantees the imaging quality and response speed of the key area, but also improves the perception ability of the monitoring system to the global anomaly and the overall monitoring breadth, and is especially suitable for scenarios of high algorithm demand under long-distance low-light conditions such as regional security monitoring and astronomical observation.

[0053] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the case of no conflict, the embodiments described below and the features in the embodiments can be combined with each other. Please refer to Figure 1 , Figure 1 is a schematic flowchart of an image processing method in a long-distance low-light scene provided by the embodiments of the present application, as Figure 1 shown, the embodiments of the present application provide an image processing method in a long-distance low-light scene.

[0054] S101, based on all pixel units of the large-aperture telescope system, initial image data is collected at a first preset period, and the initial image data is processed based on a preset image processing algorithm, and a first monitoring image is output.

[0055] Specifically, the full-view image collection is triggered according to the set first preset period, at this time, all pixel units of the large-aperture telescope system are synchronously photosensitive and output initial image data. These image data generally have problems such as serious noise, low contrast and fuzzy details due to long-distance imaging and low-light conditions. The preset image processing algorithm is called to centrally process all initial image data to suppress noise and restore key texture information, and finally generate a first monitoring image that can be used for monitoring analysis. It should be understood that the algorithm consumption is large in this process, but since it is only executed at a low frequency, the overall operation stability of the monitoring system can be guaranteed, and at the same time, a high-quality image basis is provided for the accurate identification of the subsequent area of interest.

[0056] The first preset period refers to the time interval for performing full field of view image acquisition and complete high-precision processing. The specific duration can be set comprehensively according to the field of view range, monitoring accuracy and other dimensions of the large-aperture telescope system. It should be understood that, since the large-aperture telescope system has a wide field of view, each identified interest region covers a large range in the actual geographical space, and the motion speed of a conventional monitoring target (such as a person, a vehicle or a moving celestial body in an astronomical background) is relatively limited, which belongs to a quasi-static target and is difficult to complete rapid movement across the interest region or completely leave the current interest region within the first preset period. Therefore, the first preset period can be set as a relatively long time interval under the premise of ensuring continuous monitoring, thereby reducing the algorithm burden of high-frequency full-image processing.

[0057] The preset image processing algorithm refers to a series of image processing algorithms designed for the characteristics of long-distance low-light images. The specific algorithm can be selected as needed according to related technologies.

[0058] S102, based on the first mutation standard, analyzing the brightness change in the initial image data, detecting the brightness mutation region; determining a plurality of first interest regions in the first monitoring image based on the brightness mutation region, each first interest region corresponding to a group of target pixel units.

[0059] Specifically, based on the initial image data, the brightness change is analyzed, the local brightness and the brightness difference of the surrounding neighborhood are calculated in a pixel-by-pixel or region-by-region manner, the brightness gradient or difference intensity is obtained, when the brightness gradient or difference intensity of the local image region exceeds the preset first mutation standard, it is determined that a specific event such as a moving target or a brightness mutation may occur in the local image region, and the corresponding local image region is marked as a brightness mutation region. The first interest region obtained by division includes at least one brightness mutation region, that is, multiple brightness mutation regions with close distances can be combined and divided into one interest region.

[0060] Illustratively, the image gradient is used to detect regions with significant changes in image brightness, and the spatial continuity is combined to merge multiple high-difference pixels or regions that are adjacent or less than a preset distance apart into connected regions, and filter out noise with too small area or irregular shape, thereby confirming the brightness mutation region with actual significance.

[0061] Illustratively, in the generated first monitoring image, the brightness mutation region is mapped to the first interest region, such as a rectangular region including the brightness mutation region, and each first interest region corresponds to a group of target pixel units continuously distributed on the sensor.

[0062] The first mutation criterion is a threshold criterion for judging whether there is a significant brightness change in the image. The threshold criterion can be set based on the brightness gradient and pixel difference value. The significant brightness change in the image is used as a division rule for the region of interest, which matches scenes such as the movement of a target into the field of view in a weak light background or the appearance of a sudden light source.

[0063] Further, steps S1031-S1032 are performed in the first preset period, including:

[0064] S1031, based on the available remaining computing power of the target pixel unit, selecting a supplementary monitoring area based on the remaining computing power; wherein each supplementary monitoring area corresponds to a group of monitoring pixel units;

[0065] S1032, based on the target pixel unit of the large-aperture telescope system, collecting target image data and / or monitoring pixel unit collecting monitoring image data, and based on a preset image processing algorithm, processing the target image data and / or the monitoring image data, and outputting a second monitoring image.

[0066] The remaining computing power refers to the schedulable computing resources left after completing the high-precision image processing task of the target pixel unit in the current processing period, which is used to support the monitoring task of the supplementary monitoring area. For example, the average computing power consumption of a preset image processing algorithm (such as a first processing algorithm or a second processing algorithm) when processing a unit pixel is calibrated in advance, and the computing power required for executing the first processing algorithm is estimated in combination with the current number of target pixel units, so as to obtain the schedulable computing power resources, i.e. the remaining computing power, in combination with the total system computing power. For example, according to the system computing power, the upper limit of the number of interest regions that can be processed at the same time is pre-evaluated, such as 16, and the number of supplementary monitoring areas that can be selected is determined according to the number of first interest regions, i.e. the remaining computing power.

[0067] According to the remaining computing power, one or more supplementary monitoring areas outside the first interest region are selected, for example, an image area is randomly selected each time as a supplementary monitoring area to expand the monitoring coverage or screen potential abnormalities, and each area corresponds to a group of independent monitoring pixel units. Preferably, the monitoring pixel units corresponding to the supplementary monitoring area require computing power that does not exceed the remaining computing power.

[0068] In the large-aperture telescope system, only the image data collected and output by the target pixel unit and the monitoring pixel unit is processed by using a preset image processing algorithm to output a second monitoring image. When the remaining computing power is insufficient to support the processing of a supplementary monitoring area, only the target image data collected by the target pixel unit is collected and processed, and when the first interest region is not detected, only the monitoring image data collected by the monitoring pixel unit is collected and processed.

[0069] In some embodiments, the preset image processing algorithm includes a first type of processing algorithm, which is used to process the initial image data and the target image data to output the first monitoring image or the second monitoring image.

[0070] It should be understood that the first type of processing algorithm is used to improve the image quality so that the image is clearer and easier to analyze, which can be a denoising, enhancement, reconstruction, color adjustment, or other related algorithm. Processing the initial image data based on the first type of processing algorithm ensures the initial image quality, improves the accuracy of the interest region identification, and avoids missing or misjudging the key target due to image quality. The target image data also needs to maintain a high image quality as the key focus area for subsequent monitoring analysis.

[0071] In some embodiments, the first type of processing algorithm includes a wavelet filter denoising algorithm, a multi-scale Retinex image enhancement algorithm, and an edge enhancement algorithm executed in sequence.

[0072] First, the wavelet filter denoising algorithm is executed. The signal is decomposed into wavelet coefficients of different frequencies and positions through wavelet transform, a Morlet wavelet basis function is selected for multi-resolution analysis of the signal, and the signal is localized in the time and frequency domains. Further, the signal is decomposed into different scales, different scales correspond to different frequency ranges, the low-frequency part corresponds to the overall trend and contour of the signal, and the high-frequency part corresponds to the details and noise of the signal. Since natural images have sparsity in the wavelet domain, noise usually appears as non-sparse high-frequency components, and noise and signal can be separated through threshold processing. According to different scenes, different thresholds are set. For wavelet coefficients higher than the threshold, it is presumed that they contain important information of the signal and are retained. For wavelet coefficients lower than the threshold, it is presumed that they are mainly noise or unimportant details, and they are set to zero or subjected to appropriate attenuation processing. After threshold processing, the wavelet coefficients are reconstructed using inverse discrete wavelet transform (IDWT) to obtain the filtered signal, which effectively suppresses noise while retaining the main features of the original signal.

[0073] In some embodiments, the image is decomposed into low-frequency approximation components (such as overall structure) and high-frequency detail components (such as horizontal, vertical, and diagonal edge and noise) through discrete wavelet transform (DWT), and a threshold is applied to the wavelet coefficients of the high-frequency subband to suppress the small coefficients corresponding to noise and retain the large coefficients corresponding to the signal.

[0074] In some embodiments, the threshold is dynamically adjusted according to the statistical property (e.g. variance) of the local region of the image, for example, a higher threshold is used in a region with stronger noise, and a lower threshold is used in a region with rich details. In another embodiment, the standard deviation of the noise is estimated by wavelet coefficient statistics (e.g. median method), and then the threshold parameter is adaptively adjusted according to the image content. The processed sub-band coefficients are reconstructed into the denoised image by inverse wavelet transform.

[0075] It should be understood that the distribution characteristics of noise in the dim light environment in nature are more suitable for processing by wavelet filtering, and in extremely weak light conditions, wavelet filtering can remove thermal noise in long-exposure images and enhance image details.

[0076] Secondly, the multi-scale Retinex image enhancement algorithm is executed. The Retinex algorithm separates the reflected light and incident light in the image by estimating the illumination component and the reflection component of the image, eliminates the influence of uneven illumination by processing the incident light, and enhances the details and contrast of the image. The multi-scale Retinex image enhancement algorithm processes the image at different scales to decompose the image into multiple sub-images with different resolutions. The convolution result of the small-scale Gaussian kernel is closer to the local pixel value and can capture the detailed information in the image. The illumination estimation is biased towards the high-frequency components in the image, but it may be sensitive to noise. The convolution result of the large-scale Gaussian kernel is smoother and can better reflect the overall illumination trend of the image. The illumination estimation is biased towards the low-frequency components in the image and is relatively insensitive to noise. By integrating the processing results at different scales, the image details can be enhanced while better adapting to different lighting conditions, reducing the influence of noise, and achieving a balance between detail preservation and noise suppression. In some embodiments, due to the large dynamic range of the reflection component after Retinex processing, the luminance value may exceed the display range, and normalization is required to compress the dynamic range and improve the visual contrast.

[0077] Finally, the edge enhancement algorithm is executed. Preferably, the edge enhancement algorithm is image sharpening based on HOG (Histogram of Oriented Gradients) gradient edge enhancement. Specifically, the gradient amplitude and direction of the local region of the image are calculated, and the gradient distribution histogram is constructed to extract the structural features of the image and accurately locate the edge structure. Then, the edge is weighted and enhanced using the gradient information to highlight the texture details and suppress noise interference, which is suitable for the problem of edge weakening caused by atmospheric disturbance or optical blur in long-distance imaging, especially in low-contrast low-light images, and has good detail recovery capability. In some embodiments, before executing the edge enhancement algorithm, if the image noise is high, a small-scale Gaussian kernel can be used for smoothing to suppress the interference of noise on gradient calculation. The image quality adjustment can be adjusted to avoid excessive smoothing and edge blurring.

[0078] It should be understood that in the long-distance weak light image processing, the processing sequence of first wavelet filtering denoising and then multi-scale Retinex enhancement is adopted, which improves the overall clarity and visual quality of the image. First, the high-frequency noise in the image is effectively separated and suppressed through wavelet transform, and a purer image structure is obtained before brightness correction, which reduces the risk of amplifying noise when brightening the dark area in the Retinex enhancement, and focuses on the illumination correction of the real image content, thereby realizing a more uniform dark area brightening effect and reducing the phenomenon of local dark or uneven brightness. In addition, the method of filtering first and then enhancing helps the Retinex algorithm to more accurately estimate the illumination component, avoid brightness estimation deviation caused by noise interference, and prevent the occurrence of "noise point bright spots" and other artifacts.

[0079] Please refer to Figure 2 and Figure 3 , Figure 2 is a schematic diagram of a local initial image of a first region of interest provided by an embodiment of the present application, Figure 3 is a schematic diagram of a local monitoring image of a first region of interest provided by an embodiment of the present application. As Figure 2 indicated, the shooting environment is in a deep mountain at night, and due to the long distance of the target and the dim light, the light reaching the sensor is extremely weak, and the original collected initial image has problems such as low brightness, poor signal-to-noise ratio, and blurred details. As Figure 3 indicated, after processing by a type of processing algorithm, a local monitoring image corresponding to the first region of interest is obtained, which restores the contour edge of the image, and the image quality in the monitoring image is significantly improved, which can effectively improve the accuracy of subsequent monitoring analysis.

[0080] In some embodiments, the preset image processing algorithm further includes a second type of processing algorithm, and the second type of processing algorithm is used for motion detection on the collected image data.

[0081] In some embodiments, the second type of processing algorithm includes motion detection algorithms such as inter-frame difference algorithm, background modeling algorithm, and ViBe algorithm, and each type of algorithm can be selected or used cooperatively to improve detection accuracy. The specific implementation method can be referred to related technologies. It should be understood that the background modeling algorithm can effectively distinguish between fixed environment and dynamic change by continuously learning and updating the static background model of the scene, thereby suppressing common interference such as slow change of light, movement of cloud shadow, and shaking of leaves. In long-distance weak light imaging, environmental light fluctuation is easy to cause pixel value drift, and direct use of inter-frame difference may cause false detection. After introducing the background modeling, the current image is differentiated with the background model, only the pixel area significantly deviating from the background is retained, and the influence of light interference is reduced. On this basis, combined with the inter-frame difference algorithm, the continuous displacement characteristics of the moving target can be further detected, and stable and robust motion detection can be realized with low algorithm power consumption.

[0082] It should be noted that the processing target and computational complexity of the first type of processing algorithm and the second type of processing algorithm are different, so that the required computing power of the first type of processing algorithm is greater than that of the second type of processing algorithm. The first type of processing algorithm such as wavelet filter denoising, multi-scale Retinex enhancement and HOG-based edge enhancement needs to perform multi-scale decomposition, nonlinear transformation and local refinement processing on images with high noise and low signal-to-noise ratio, involving complex mathematical operations and multiple iterations to achieve strong denoising, dark area brightening and edge sharpening effects. Especially on high-resolution and large-field images generated by large-aperture telescopic systems, the number of pixels is huge, and such algorithms need to be calculated intensively in the full image or large area, resulting in high computing power consumption. The second type of processing algorithm such as inter-frame difference and background modeling is mainly used for fast detection of image changes, only needs to perform simple difference comparison or threshold judgment on adjacent frame pixel values, has small amount of calculation and fast response speed, and does not involve complex image reconstruction process, which is a lightweight algorithm. Therefore, the first type of processing algorithm improves the image quality while bringing higher computing power requirement, while the second type of processing algorithm realizes fast abnormal preliminary screening with low overhead, and the cooperation of the two can adapt to the actual running environment with limited computing power while ensuring the imaging quality of the core area.

[0083] In some embodiments, the division of the region of interest is realized by the region of interest (ROI) mode based on the HOG gradient edge enhancement algorithm. Specifically, by analyzing the first monitoring image through the edge enhancement algorithm based on the HOG gradient feature, the region where the target with clear contour or geometric feature is located can be effectively identified, and it is divided into a first region of interest or a second region of interest. When the first monitoring image performs the edge enhancement step in the first type of processing algorithm, if it itself adopts the same image sharpening based on the HOG gradient edge enhancement, it can be judged that the processing has been completed in advance, and there is no need to repeat the calculation, and the gradient enhancement result or edge map data generated previously can be directly reused, skipping the repeated processing link.

[0084] The image processing in the first preset period enters the ROI mode, a local region on the CMOS image sensor is defined as an ROI region (such as a first interest region or a supplementary monitoring region), only the pixels in the region are subjected to light sensing and signal processing, and the pixels outside the region are ignored, thereby cropping the light sensing region, simplifying the readout process, and improving the real-time performance of image processing and output. Through the driving circuit of the sensor, only the pixel units in the ROI region are activated for exposure and charge accumulation, data is subjected to high-resolution sampling, enhancement processing or real-time analysis (i.e., a preset image processing algorithm); the pixels in the non-ROI region do not participate in imaging, and the resolution or compression thereof is reduced, thereby reducing the data volume. The pixel data in the ROI region can be directly transmitted to the processor through the high-speed readout circuit, skipping the data reading of the non-ROI region, reducing redundant processing, and performing real-time monitoring on the scene according to the output monitoring image. When a specific event (such as motion or brightness mutation) occurs in the ROI region, recording or alarm is started.

[0085] It should be understood that the ROI mode only selects the pixels in a certain rectangular region to participate in light sensing and signal output, and the pixels in the remaining region do not work or the signal is discarded, which can reduce the image information amount transmitted by the camera and the image information amount required to be processed by the computer, thereby improving the analysis efficiency. Moreover, the algorithm can be more focused on the key part of the image, unnecessary data interference is avoided, and the processing quality and decision accuracy are improved.

[0086] In some embodiments, since the ROI mode directly intercepts a part on the original chip, the field of view and the light sensing property are correspondingly reduced. Assuming that only 1 / 4 of the pixels participate in light sensing, since the light sensing area (pixel size) of a single pixel is fixed, the total light sensing area is directly reduced to 1 / 4 of the original, and the total amount of photons collected by the sensor per unit time is reduced by 75%. To alleviate the dark state problem of the ROI region, the exposure time can be prolonged, and more light sensing time can be allocated to make up for the insufficient amount of collected photons.

[0087] In some embodiments, the S1032 comprises: based on the second processing algorithm, performing motion detection according to historical monitoring images and the monitoring image data, and determining whether there is an abnormality in the supplementary monitoring region; if yes, updating the second monitoring image based on the processing of the monitoring image data in the supplementary monitoring region according to the first processing algorithm; or performing steps S101 to S102.

[0088] Specifically, historical image data of the supplementary monitoring area in the historical monitoring image is acquired, current image data and the historical image data of the supplementary monitoring area are analyzed by a two-class processing algorithm, and a brightness change in consecutive frames is determined as an anomaly if the brightness change exceeds a normal fluctuation range (such as a pre-set threshold). The historical image data is a pre-set number of image frames before the current time, and the specific value can be flexibly set according to requirements, for example, 3-5 consecutive image frames. For example, when a moving figure appears in the supplementary monitoring area, even if the overall image is blurred and the signal-to-noise ratio is low due to long-distance weak light imaging, the figure will continuously block or reflect the ambient light during the movement, causing a detectable brightness change of the local area where the figure is located relative to the surrounding static background (such as the ground, vegetation, or buildings). This change is manifested as a fluctuation or displacement trajectory of the brightness value in consecutive frames, and even if the target outline is not clear, it can still be captured by a two-class processing algorithm such as frame difference or background modeling for motion detection.

[0089] When an anomaly is detected, a subsequent high-precision processing or global reanalysis process is triggered according to the range and characteristics of the anomaly, realizing early detection and confirmation of potential intrusion behavior.

[0090] Optionally, if the data anomaly is relatively local and the shape is clear, the monitoring image data corresponding to the supplementary monitoring area is processed based on the one-class processing algorithm, taking into account efficiency and imaging quality. For example, a single moving figure or vehicle appears in a supplementary monitoring area of a security area, and a one-class processing algorithm can be called to process the image data of the local area with high precision, restore the target outline and details through wavelet denoising, Retinex enhancement, and other means, and fuse the enhanced results into the current second monitoring image, realizing the quality improvement of the key area without interrupting the overall monitoring process.

[0091] Optionally, if the data anomaly covers a large area or has a complex shape, such as multiple data anomalies in the supplementary monitoring area, or the data anomaly area in the supplementary monitoring area is greater than a pre-set area, there may be a sudden large-area interference or potential multi-target activity in the scene, and the like. In the next processing cycle, S101 to S102 are executed, the full-view image is re-acquired, and the complete interest area recognition is performed to re-evaluate the scene state from a global perspective, avoiding false alarms or misjudgments caused by local misjudgments, and improving the reliability of the monitoring system.

[0092] In some embodiments, when the remaining computing power is sufficient (such as when the number of target pixel units is small), both the target pixel units and the monitoring pixel units can be uniformly processed by a one-class processing algorithm, and the final output is a second monitoring image including refined target image data and monitoring image data.

[0093] In some embodiments, when the remaining computing power is limited, the target pixel unit and the monitoring pixel unit are processed by a first processing algorithm and a second processing algorithm respectively, and the final output second monitoring image can include the target image data after fine processing and part of the monitoring image data, the target image data after fine processing and the monitoring image data, or the target image data after fine processing and all the monitoring image data.

[0094] In some embodiments, the method comprises: obtaining a plurality of preset second regions of interest, the second regions of interest being high-risk regions where key monitoring objects or potential objects with sudden changes in brightness exist; and polling at least one second region of interest as a supplementary monitoring region from the plurality of second regions of interest at a second preset period.

[0095] Specifically, a plurality of second regions of interest are pre-configured, such as entrances and exits of security areas, commanding heights, or known variable star regions in astronomical observations, which can be high-risk positions where key monitoring objects or sudden changes exist. The plurality of second regions of interest can be stored in a sequence or a priority list, and are iterated at a second preset period. At least one region is selected as a target second region of interest for supplementary monitoring in each second preset period, for example, if there are four second regions of interest, they are rotated every three seconds, achieving periodic coverage.

[0096] After the second region of interest is selected as the supplementary monitoring region, the corresponding monitoring pixel unit collects monitoring image data by photosensitive sampling, and outputs the data by the first processing algorithm or performs motion detection by the second processing algorithm, so that the key region can be continuously focused on and potential threats or events can be discovered in time even when the computing power is limited. In some embodiments, one or more second regions of interest can be selected as the supplementary monitoring region according to the remaining computing power.

[0097] In some embodiments, the binary processing algorithm adaptively adjusts according to a second preset period under the polling mechanism. First, a short-term frame buffer mechanism is established for each second region of interest (ROI) to store the monitoring image data corresponding to each ROI, serving as cached historical image data for later use. Since polling is performed periodically, when a ROI is selected, the latest image data at the current moment is read and compared with the cached historical image data. Motion detection is then performed using inter-frame differencing and / or background modeling methods. For example, at the 2nd second when ROI A is selected, motion detection is performed on the current image data of the current frame and the historical image data from the previous one or more polls of ROI A, based on the binary processing algorithm. The cached historical image data is synchronously refreshed with each polling update, ensuring that each motion detection is based on the most recent image data for that ROI.

[0098] The second preset period refers to the time interval for polling and monitoring the second region of interest. It is set to be shorter than the first preset period to achieve high-frequency, periodic scanning of high-risk areas. This ensures that key areas can be polled multiple times between two full-view processing sessions, thereby improving the sensitivity to responding to abnormal events.

[0099] In some embodiments, the method includes: pre-marking several second regions of interest. Several key areas, such as entrances and exits of walls or specific areas of the sky in astronomical observations, can be pre-marked within the field of view as fixed monitoring focuses.

[0100] In some embodiments, the method includes: identifying at least one key monitoring object from the first monitoring image based on a preset target recognition algorithm, and designating the image region where the key monitoring object is located as a second region of interest. The preset target recognition algorithm refers to a lightweight recognition method for detecting specific target categories from an image, such as a classifier based on template matching or a shallow neural network, used to locate the key monitoring object; specific implementations can be found in related technologies. Correspondingly, the key monitoring object refers to a target entity in the monitoring scene that requires focused attention, such as personnel, vehicles, or specific astronomical phenomena.

[0101] In some embodiments, the method includes: analyzing brightness changes in the initial image data based on a second mutation criterion, and identifying potential brightness mutation regions as second regions of interest; wherein the index requirements in the second mutation criterion are lower than those in the first mutation criterion.

[0102] The second mutation criterion is a threshold criterion for detecting potential brightness changes, which can be set based on brightness gradients and pixel difference values, and has a threshold lower than that of the first mutation criterion, for identifying suspicious areas that have not yet reached significant brightness changes but have brightness changes, such as weak or gradual brightness changes. For example, in night monitoring, a certain area has a slight brightness fluctuation, which is identified as a suspicious area, and the suspicious area is mapped as a second area of interest, such as a rectangular range containing the suspicious area, which participates in subsequent polling monitoring, improving the early warning capability.

[0103] In some embodiments, the S1031 comprises: analyzing the brightness change rule of the first area of interest based on historical monitoring images, identifying false areas of interest in the first area of interest based on the brightness change rule; polling and selecting valid false areas of interest from the false areas of interest at a third preset period, and updating the first area of interest; wherein the third preset period is less than the first preset period.

[0104] Specifically, by analyzing the brightness change rule of the first area of interest in the historical monitoring image, the area with periodic or predictable change characteristics is identified, which frequently triggers the brightness mutation detection but has no actual monitoring value, and is determined as a false area of interest. The false area of interest is a invalid first area of interest by default, and the corresponding pixel unit does not perform photosensitive and data output, and does not participate in subsequent data processing. It should be understood that the false area of interest corresponds to a scene where brightness changes repeatedly due to environmental interference (such as vegetation shaking, water surface reflection, and cloud movement).

[0105] In each third preset period, one or more are updated to valid first areas of interest are selected in order from all false areas of interest and high-quality images are output, and in subsequent monitoring analysis, it can be verified whether there is an abnormality in the false area of interest, for example, checking whether a target crosses the tree shadow area. If no abnormal change occurs, the attribute of the false area of interest is maintained; if an abnormal change is detected, it is updated to a fixed first area of interest. The state of the false area of interest is dynamically verified in a low-frequency polling manner to prevent false exclusion, while releasing computing power for monitoring of other key areas, to some extent, so that part of the supplementary monitoring area can also be processed by a type of processing algorithm, and high-quality images are output.

[0106] The third preset period refers to the time interval for polling and reviewing the false area of interest, which is set to be shorter than the first preset period, so that high-quality images of the false area of interest are output in the first preset period, avoiding missing key monitoring objects and improving monitoring robustness.

[0107] It should be understood that the embodiments of the present application dynamically update the monitoring area through a polling mechanism. The formal interest area is identified through the first monitoring image, and the periodic polling and updating of the second interest area and the pseudo interest area ensure both continuous attention to known key targets and identification and adaptation to new scene changes.

[0108] Referring to Figure 4 , Figure 4 is a running flowchart of a monitoring imaging system provided by the embodiments of the present application.

[0109] As Figure 4 shown, in the image acquisition process, light is collected through the LX90-ACF astronomical telescope to improve light flux and reduce the influence of noise on imaging, thereby improving the sensitivity and resolution of imaging; the AR sheet is customized for the visible light wavelength band; the micro-lens array focuses the incident light onto the photodetector to improve the light collection efficiency; the light collected after the micro-lens array is irradiated onto the image sensor of the CMOS camera (i.e., the preset position), converts the light signal into an electrical signal, and performs preliminary photoelectric conversion and signal sampling, thereby converting the light information into digital image data. The CMOS camera has three operating modes, namely, a global shutter (e.g., based on the initial image data collected by all pixel units of the large-aperture telescope system), parallel A / D conversion, and ROI mode (e.g., based on the target image data collected by the target pixel unit and / or the monitoring image data collected by the monitoring pixel unit of the large-aperture telescope system).

[0110] Further, the data is processed through the ISP chip integration algorithm, which is used to implement a class of processing algorithms, such as wavelet filtering (i.e., wavelet filtering denoising algorithm), multi-scale Retinex low-illumination image enhancement (i.e., multi-scale Retinex image enhancement algorithm), and image sharpening based on HOG gradient edge enhancement (i.e., edge enhancement algorithm). At the same time, the device is connected to a client with display and data processing functions such as a computer, and the image is displayed in the preset special software, manual adjustment of focusing is supported to update the imaging parameters of the telescope, and adjustment of exposure, gain, and other parameters in the software is also supported. After the user adjusts the parameters, the feedback can be fed back to the ISP chip to dynamically update the integrated algorithm, and finally the imaging under extremely weak light conditions is completed.

[0111] In some embodiments, the ISP chip, i.e., the image signal processing chip, converts the original image signal into a digital signal through AD conversion, and then performs memory storage and buffering processing, so as to perform subsequent image processing. For the collected digital image signal, various processing and optimization are performed according to the preset algorithm, such as automatic adjustment of exposure, gain, and multiple parameters, to maximize the image quality, and finally output the image format suitable for image display devices.

[0112] The monitoring image (such as the first monitoring image and the second monitoring image) obtained by the embodiment of the application can improve the human eye recognizable degree, and provides high-quality input for subsequent target detection, recognition and behavior analysis algorithms, realizes continuous observation and safety warning of a specific scene, and is widely applied in security, transportation, astronomical observation, environmental monitoring and other fields. The monitoring image can be used to find suspicious stay, entry and exit behaviors in a security area, support real-time warning and command decision; in astronomical observation, the monitoring image can be used to capture dark and weak celestial bodies, moving celestial bodies or transient astronomical events, and assist scientific research.

[0113] Please refer to Figure 5 , Figure 5 is a structural schematic block diagram of a computer device provided by the embodiment of the application. The computer device can be a terminal device or a server.

[0114] Exemplarily, the method described above can be implemented in the form of a computer program, which can run on the computer device as shown in Figure 5 .

[0115] As shown in Figure 5 , the computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0116] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which when executed, can make the processor execute any kind of image processing method in a long-distance weak light scene.

[0117] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0118] The internal memory provides an environment for the running of the computer program in the non-volatile storage medium, which when executed by the processor, can make the processor execute any kind of image processing method in a long-distance weak light scene.

[0119] The network interface is used for network communication, such as sending assigned tasks.

[0120] It should be appreciated that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0121] In one embodiment, the processor is configured to execute a computer program stored in the memory to implement the following steps:

[0122] S101, acquiring initial image data based on all pixel units of the large-aperture telescope system at a first preset period, and processing the initial image data based on a preset image processing algorithm to output a first monitoring image;

[0123] S102, analyzing a brightness change in the initial image data based on a first mutation criterion to detect a brightness mutation region, and determining a plurality of first regions of interest in the first monitoring image based on the brightness mutation region, each first region of interest corresponding to a group of target pixel units;

[0124] Steps S1031-S1032 are performed within the first preset period, including:

[0125] S1031, selecting a supplementary monitoring region based on the remaining computing power available based on the target pixel units, wherein each supplementary monitoring region corresponds to a group of monitoring pixel units;

[0126] S1032, acquiring target image data based on the target pixel units of the large-aperture telescope system and / or monitoring image data based on the monitoring pixel units, and processing the target image data and / or the monitoring image data based on a preset image processing algorithm to output a second monitoring image.

[0127] For example, the processor is configured to execute a computer program stored in the memory, and is further configured to implement the steps of the image processing method in a remote weak light scene according to any embodiment of the present application, which will not be described here.

[0128] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program comprises program instructions. The processor executes the program instructions to realize the steps of the image processing method in the long-distance weak light scene according to any one of the embodiments of the present application.

[0129] The computer readable storage medium can be an internal storage unit of the computer device, for example, a hard disk or a memory of the computer device. The computer readable storage medium can also be an external storage device of the computer device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0130] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for image processing in a distant low-light scene, characterized in that, The method is applied to a large-aperture telescopic system, and the method comprises: S101. In a first preset period, initial image data is collected based on all pixel units of the large-aperture telescopic system, and a first monitoring image is output by processing the initial image data based on a preset image processing algorithm; S102. Based on a first mutation standard, a brightness change in the initial image data is analyzed to detect a brightness mutation region; a plurality of first interest regions are determined in the first monitoring image based on the brightness mutation region, and each first interest region corresponds to a group of target pixel units; Steps S1031-S1032 are performed in the first preset period, comprising: S1031. Based on the remaining computing power available for the target pixel units, a supplementary monitoring region is selected based on the remaining computing power; wherein each supplementary monitoring region corresponds to a group of monitoring pixel units; S1032. Target image data is collected based on the target pixel units of the large-aperture telescopic system and / or monitoring image data is collected based on the monitoring pixel units, and a second monitoring image is output by processing the target image data and / or the monitoring image data based on a preset image processing algorithm; The preset image processing algorithm comprises a first type of processing algorithm and a second type of processing algorithm, the first type of processing algorithm is used to process the initial image data and the target image data; the computing power required by the first type of processing algorithm is greater than the computing power required by the second type of processing algorithm; S1032 comprises: based on the second type of processing algorithm, motion detection is performed according to historical monitoring images and the monitoring image data to determine whether there is an anomaly in the supplementary monitoring region; if so, the second monitoring image is updated based on the processing of the monitoring image data of the corresponding supplementary monitoring region by the first type of processing algorithm; or steps S101 to S102 are performed; A plurality of second interest regions are obtained, the second interest regions are high-risk regions in which key monitoring objects or potential brightness mutation objects exist; at least one second interest region is selected as a supplementary monitoring region from the plurality of second interest regions in a second preset period; wherein the second preset period is less than the first preset period.

2. The method of claim 1, wherein, The large-aperture telescopic system comprises a large-aperture telescope with a preset resolution and an AR sheet, a microlens array, and a CMOS image sensor in a preset waveband; The large-aperture telescope comprises a concave primary mirror, a plane secondary mirror, and an eyepiece, a lens group is arranged between the concave primary mirror and the eyepiece to form a relay surface structure, so as to expand the imaging range by focusing the eyepiece; The AR sheet in the preset waveband is arranged in front of the microlens array and is used for filtering incident light; The microlens array is used for redirecting or focusing incident light to a corresponding preset position; The CMOS image sensor is used for photosensing all or part of the pixel units to obtain corresponding image data.

3. The method of claim 1, wherein, The first type of processing algorithm comprises wavelet filtering denoising algorithm, multi-scale Retinex image enhancement algorithm, and edge enhancement algorithm which are executed in sequence; the second type of processing algorithm comprises inter-frame difference algorithm and / or background modeling algorithm.

4. The method of claim 1, wherein, The method comprises: Calibrate a plurality of second interest regions in advance; or Identify at least one key monitoring object from the first monitoring image based on a preset target recognition algorithm, and take the image region where the key monitoring object is located as a second interest region; or Analyze the brightness change in the initial image data based on a second mutation standard, and take the detected potential brightness mutation region as a second interest region; wherein the index requirement of the second mutation standard is lower than that of the first mutation standard.

5. The method of claim 1, wherein, The S1031 comprises: Analyze the brightness change rule of the first interest region based on the historical monitoring image, and identify the false interest region in the first interest region based on the brightness change rule; Poll the effective false interest region from a plurality of false interest regions at a third preset period, and update the first interest region; wherein the third preset period is less than the first preset period.

6. A computer device, comprising: The device comprises: A memory for storing a computer program; A processor for executing the computer program and implementing the image processing method in a long-distance weak light scene as claimed in any one of claims 1 to 5 when executing the computer program.

7. A computer readable storage medium characterized by The computer readable storage medium stores a computer program, and the computer program makes the processor implement the image processing method in a long-distance weak light scene as claimed in any one of claims 1 to 5 when executed by the processor.

Images