An Image Processing Method and System Applicable to Pod Edge Computing Devices

Abstract

The present invention discloses an image processing method and system applicable to a pod edge computing device. The method includes: collecting an image; performing gray level equalization on the collected image to obtain an image after gray level equalization; filtering the image after gray level equalization to obtain a filtered image; and performing motion compensation on the filtered image to obtain a final image. The system includes: a collection module, an equalization module, a filtering module, and a compensation module. The present invention eliminates the trailing and blurring phenomena of moving targets in the original image collected by an optoelectronic pod through motion compensation, and provides a feasible solution for the application of the infrared image preprocessing technology of the optoelectronic pod. The present invention can be widely applied to the field of image processing.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular, to an image processing method and system applicable to pod edge computing devices. Background Art

[0002] An optoelectronic pod utilizes optoelectronic sensors in different bands such as visible light, infrared, laser, and millimeter wave, and through a precise stabilization platform and a high-speed airborne control system, realizes functions such as search, tracking, identification, and measurement of land, sea, and air targets.

[0003] Generally speaking, due to the special properties of infrared images collected by optoelectronic pods, such as low heat distribution and reflectivity, it is difficult to directly analyze and process them. Therefore, in order to improve the clarity and utilization efficiency of infrared images, infrared image preprocessing technology has become an indispensable part of optoelectronic pods.

[0004] When an optoelectronic pod collects infrared images, it is easy to occur that the target moves at a high speed during the photography process, the exposure time is relatively long, when the target rapidly changes temperature, in low light or poor lighting conditions, etc., resulting in the target may leave a blurred or trailing effect on the image. Aiming at the infrared imaging characteristics in optoelectronic pods, the current single image processing algorithm cannot meet the requirements of its quality and real-time performance. Summary of the Invention

[0005] In view of this, in order to solve the technical problems that the existing infrared image processing methods cannot simultaneously meet the imaging quality and real-time performance, and cannot be applied to the optoelectronic pod application scenario, on the one hand, the present invention proposes an image processing method applicable to pod edge computing devices, and the method includes the following steps:

[0006] Collect an image;

[0007] Perform gray level equalization on the collected image to obtain an image after gray level equalization;

[0008] Based on an adaptive filtering window size, filter the image after gray level equalization to obtain a filtered image;

[0009] Perform motion compensation on the filtered image to obtain a final image.

[0010] Among them, the pod-mounted device is remotely controlled by a mobile device to move, and an image acquisition instruction is sent to the cloud intelligent control computing center. The cloud dispatches a collaborative work task to the edge computing device according to the user instruction, and the edge computing device collects infrared image data according to the task content.

[0011] In some embodiments, it further includes:

[0012] Perform preliminary recognition and region division on the acquired image to obtain image blocks of different regions.

[0013] In some embodiments, the step of performing preliminary recognition and region division on the acquired image to obtain image blocks of different regions specifically includes:

[0014] Process the acquired image based on the Canny edge detection algorithm and the object detection algorithm to obtain region data and edge data;

[0015] Calculate the feature point differences based on the region data and the edge data to obtain image blocks of different regions.

[0016] In some embodiments, the formula for gray level equalization is expressed as follows:

[0017]

[0018]

[0019]

[0020] Among them, G1(i, j) is the gray value of the pixel points in G k (i, j) in the image processing window that satisfy |G1(i, j) - GE last | ≤ G delta ; G delta is the binding parameter; G2(i, j) is the gray value of the pixel points in G k (i, j) in the image processing window that satisfy |G1(i, j) - G mean_last | > G delta ; N1 is the number of pixel points in G k (i, j) in the image processing window that satisfy |G1(i, j) - G mean_last | ≤ G delta ; N2 is the number of pixel points in G k (i, j) in the image processing window that satisfy |G1(i, j) - G mean_last | > G delta ; i and j respectively represent the row index and column index of the pixels in the image, R end represents the ending row number of the processing window, R start represents the starting row number of the processing window, C end represents the ending column number of the processing window, C start represents the starting column number of the processing window.

[0021] In some embodiments, the step of filtering the gray level equalized image to obtain the filtered image specifically is:

[0022] Calculate the difference between the gray value of the original pixel point in the image after gray level equalization and the median gray value;

[0023] Determine the filtering window size according to the difference;

[0024] Perform filtering based on the filtering window size to obtain a filtered image.

[0025] In some embodiments, the formula for filtering is as follows:

[0026]

[0027] Among them, median(·) represents taking the median value, U represents the full enumeration of data within a preset range around (i,j), and m and n represent the neighborhood indices relative to the current pixel point (i,j).

[0028] In some embodiments, it further includes:

[0029] Implement parallel filtering through a multi-level pipeline.

[0030] In some embodiments, for the step of performing motion compensation on the filtered image to obtain a final image, it specifically includes:

[0031] Based on the filtered image, calculate the gray value difference between two consecutive frames of images;

[0032] Adjust the estimated energy accumulation coefficient according to the gray value difference to obtain an adjusted energy accumulation coefficient;

[0033] Use the second filtering coefficient to filter the target gray value energy of the previous frame image to obtain compensated gray value energy;

[0034] Convert the compensated gray value energy into image gray value data to obtain a final image.

[0035] In some embodiments, the formula for motion compensation is expressed as follows:

[0036]

[0037]

[0038] C k (i,j) = {f1×GF t_k (i,j)+f2×GF t_k (i - δ y ,j + δ z )+f3×GF t_k (i + δ y ,j - δ z )+f4×GF t_k(i - δ y , j - δ z )};

[0039]

[0040] Among them, γ1 represents the first filtering coefficient, γ2 represents the second filtering coefficient, represents the gray value of the previous frame image at position (i, j); GF t_k (i, j) represents the gray difference between two consecutive frames of images, represents the gray value of the current frame image at position (i, j); GF k (i, j) represents the gray value of the current frame after filtering at position (i, j); δ γ represents the estimated value of the displacement of the image in the y direction; δ z represents the estimated value of the displacement of the image in the z direction; C k (i, j) represents the compensated gray energy; (i, j) is I k (i, j) is the average of the absolute values of I

[0041] The present invention also proposes an image processing system applicable to a pod edge computing device, and the system includes:

[0042] An acquisition module for acquiring images;

[0043] An equalization module for performing gray equalization on the acquired images to obtain images after gray equalization;

[0044] A filtering module for filtering the images after gray equalization based on an adaptive filtering window size to obtain filtered images;

[0045] A compensation module for performing motion compensation on the filtered images to obtain final images.

[0046] Based on the above solution, the present invention provides an image processing method and system applicable to pod edge computing devices. By motion compensation, the trailing and blurring phenomena of moving targets in the original image collected by the optoelectronic pod for infrared image acquisition are eliminated, and the contrast between the moving target and the background is enhanced. In addition, to achieve motion compensation, the gray levels of the previous and current frames of images are required. According to the relative motion characteristics of the infrared target and the background, the present invention calculates the gray compensation weight after motion through a global motion estimation algorithm, and realizes the enhancement of the gray level of the target area after superposition during the superposition of the gray levels of the two frames of images. To eliminate the trailing and blurring phenomena of moving targets, weakening processing must be performed on the previous gray level image. That is, equalize the target and the background of the previous frame to make it a relatively pure background floor, and on this basis, superimpose the enhanced gray level image of the target after motion compensation to achieve the best effect, providing a feasible solution for the application of the infrared image preprocessing technology of the optoelectronic pod. Description of the Drawings

[0047] Figure 1 is the flowchart of the steps of an image processing method applicable to pod edge computing devices of the present invention;

[0048] Figure 2 is the block diagram of the structure of an image processing system applicable to pod edge computing devices of the present invention. Detailed Embodiments

[0049] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with 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. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0050] It should be noted that for the sake of convenience of description, only the parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0051] It should be understood that the "system", "device", "unit" and / or "module" used in the present application are a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, they can be replaced by other expressions.

[0052] Unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. An element defined by the statement "comprising one..." does not exclude the existence of other identical elements in the process, method, product, or device that includes the element.

[0053] In the description of the embodiments of the present application, "a plurality" means two or more than two. The following terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined by "first" and "second" may explicitly or implicitly include one or more of such features.

[0054] In addition, flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the previous or subsequent operations do not necessarily need to be executed precisely in sequence. On the contrary, the steps can be processed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0055] Referring to Figure 1 , which is a schematic flowchart of an optional example of the image processing method applicable to the pod edge computing device proposed by the present invention. This method can be applicable to a computer device. The image method proposed in this embodiment may include but is not limited to the following steps:

[0056] Step S1: Collect an image;

[0057] Step S2: Perform gray level equalization on the collected image to obtain the gray level equalized image;

[0058] Step S3: Filter the gray level equalized image based on the adaptive filtering window size to obtain the filtered image;

[0059] Step S4: Perform motion compensation on the filtered image to obtain the final image.

[0060] In some feasible embodiments, step S1 further includes:

[0061] Perform preliminary recognition and region division on the collected image to obtain image blocks of different regions.

[0062] Specifically, the acquired image is processed based on the Canny edge detection algorithm and the object detection algorithm to obtain region data and edge data; based on the region data and the edge data, the feature point differences are calculated to obtain image patches of different regions.

[0063] Through this preferred embodiment, the image can be preliminarily divided into regions, which helps to construct different filter window sizes according to different regional images subsequently.

[0064] In some feasible embodiments, the step S1:

[0065] S1.1. Use a high-resolution infrared camera: The infrared camera in the optoelectronic sensor of the optoelectronic pod consists of near-infrared, short-wave infrared, mid-infrared, and long-wave infrared.

[0066] The infrared camera carried on the optoelectronic pod can sense and capture the infrared radiation emitted by the target object, and achieve the imaging and analysis of the thermal energy distribution.

[0067] The short-wave infrared camera in the optoelectronic pod of the present invention selects an MV-E150-NIR gigabit Ethernet industrial camera. This camera has characteristics such as high definition and low noise, and the highest resolution can reach 1379×1104. It can not only clearly image in visible light, but also has better induction in the near-infrared band of 760nm - 1000nm, and is especially suitable for the optoelectronic pod tasks of near-infrared band visual detection.

[0068] S1.2. Install and position the camera: Install the camera on the optoelectronic pod, and determine its viewing range and field of view by adjusting the angle and position. The position and angle of the camera should be adjusted according to the specific application situation so as to be able to comprehensively capture the image information of the pod operation area.

[0069] S1.3. Connect and configure the camera: Integrate and install the camera into the optoelectronic pod, and perform corresponding configurations as needed. The configurations include setting parameters such as image resolution, frame rate, exposure time, and sensor gain.

[0070] S1.4. Perform image acquisition: Once the camera is connected and configured, the user can start remotely controlling the movement of the pod-mounted device through a mobile device and send commands. The camera will regularly generate image frames and transmit them through digital signals. The edge computing device of the optoelectronic pod will receive and process these image frames, thereby realizing the real-time monitoring of the pod operation area.

[0071] Through the above steps, the optoelectronic pod can achieve image acquisition of the operation area. The optoelectronic pod is a camera device loaded on a robotic arm or a jib crane, and is used for monitoring and controlling the pod operation. In the optoelectronic pod, image acquisition is carried out through a camera.

[0072] In some feasible embodiments, the step S2 specifically includes:

[0073] Grayscale equalization is performed on the infrared grayscale image of the previous frame. Contrary to the processing methods such as histogram equalization and grayscale stretching, grayscale equalization is mainly to use the grayscale mean to equalize the target and background grayscale in the image, which is to weaken the contrast of the grayscale image of the previous frame. The algorithm principle of mean filtering can be used for reference, because mean filtering suppresses the non-uniform noise of the original image, which will make the details or edges of the image blurred. If the sliding window is not designed, the grayscale statistical mean algorithm is directly used for the entire frame image to weaken the target background contrast. The effect of grayscale equalization is similar to suppressing high-frequency noise while filtering out the target, and producing a grayscale balanced, pure base suitable for subsequent motion compensation.

[0074] The formula for grayscale equalization is as follows:

[0075]

[0076]

[0077]

[0078] In some feasible embodiments, the step S3 specifically includes:

[0079] S3.1, calculating the difference between the grayscale of the original pixel and the grayscale median in the grayscale equalized image;

[0080] S3.2. Use the difference to replace the original pixel grayscale in the grayscale-equalized image to obtain a filtered image.

[0081] In this embodiment, the previous frame image can be regarded as a pure background base, and the current frame needs to eliminate the influence of background noise through filtering. On this basis, motion compensation is performed on the target area of the current frame to enhance the grayscale of the target after movement. The highly enhanced target grayscale image after movement is superimposed on the pure base to obtain an ideal high-contrast infrared image of the moving target. First, the current frame grayscale image needs to be de-noised to avoid noise from being introduced into the superimposed image output. The motion compensation process needs to enhance the grayscale of the target area after movement. By improving the traditional median filtering algorithm, the difference between the original pixel grayscale and the median is used to replace the original pixel grayscale to obtain a low-contrast filtered image. This can greatly remove high-frequency noise and edge interference, and can also ensure that the junction edge between the target and the background is clear, and the image will not become blurred while filtering out noise.

[0082] The key to median filtering lies in the selection of window size. This system uses 5*5 large template median filtering. The filtering algorithm is as shown in formula (4):

[0083]

[0084] That is, if the image size is 255*256, the spatial filtering result (pixel - median filtering output) is 251*252. Considering that the overall gray value of the image after median filtering is relatively low and the image background will tend to be black, the two rows and two columns at the image edge are regarded as blind pixels, and all blind pixels are filled with gray value 0 to compensate for the edge rows and columns that have not undergone spatial domain filtering and the over - bright and over - dark points in the image. After filling, the image size is restored to 255*256.

[0085] In addition, in some feasible embodiments, step S3 specifically includes:

[0086] S3.1. Calculate the difference between the gray value of the original pixel point in the gray - level equalized image and the gray - level median.

[0087] S3.2. Determine the filtering window size according to the difference.

[0088] S3.3. Perform filtering based on the filtering window size to obtain the filtered image.

[0089] Since the filtering window size in the filtering algorithm is mostly set according to manual experience and cannot adaptively adjust the window size and standard deviation for different regions in the image, when removing noise, some detail information will be lost. Therefore, in this preferred embodiment, the filtering window size is set using the gray - level difference, and this setting relationship can specifically be the prior data obtained through a large number of repeated experiments to achieve adaptive adjustment, which helps to retain detail information.

[0090] In some embodiments, it further includes:

[0091] Implement parallel filtering through a multi - level pipeline.

[0092] In order to give full play to the parallel computing advantage of the edge computing device, we adopt the computing mode of the optoelectronic pod edge computing device in this real - time system. This system has very high requirements for real - time performance. To meet this requirement, we design a fast matrix sorting algorithm for 5×5 median filtering to replace the value comparison in the traditional bubble method. For 5×5 median filtering, it is divided into 4 parallel filterings (b - e).

[0093] Traditional serial methods require 24 value comparisons. However, the parallel algorithm used in this patent can run through a 4-stage pipeline on edge computing devices, and its running time is only 1 / 6 of that of the serial method. Based on the computing mode of the edge computing device of the optoelectronic pod, we set the basic clock cycle of the module to 90 MHz. Taking 5×5 as an example, using this parallel algorithm can save 0.05 us compared with the traditional serial method. In the 251×252 sliding windows of the entire image, using the parallel algorithm can save 3.162 ms. Each parallel filter adopts the computing architecture of the edge computing device of the optoelectronic pod and is implemented through the parallel switching of a 5×5 matrix. The first three filters achieve filtering in all directions in the figure by changing the arrangement order of the input pixels. The fourth filter realizes the median calculation of three pixels, and by subtracting the result of the median filter from the original value of the pixel, the result of spatial filtering is obtained. In the computing mode of the edge computing device of the optoelectronic pod, 5 rows of data can be buffered first to achieve the parallel output of 25 data to be sorted. Then the sorting of 25 data is converted into the parallel sorting of 5 groups of 5 points. The sorting of 25 points is completed by using a 4-stage pipeline, and the 5 points in each group are processed in the way of matrix parallel sorting. Based on the computing mode and logic of the edge computing device of the optoelectronic pod, such parallel filtering can achieve efficient image processing.

[0094] In some feasible embodiments, step S4 specifically includes:

[0095] The motion compensation algorithm needs to achieve the target gray-scale enhancement of the current frame gray-scale image. When enhancing the target gray-scale, in order to remove the target image trailing phenomenon caused by motion, a time-domain filtering algorithm is required, that is, α-β filtering filters the images before and after compensation in the time domain and the energy domain respectively. In the time domain, the gray-scale difference is obtained by the α coefficient for the gray-scales of the previous and current frames, and the target motion characteristics are analyzed. The relative motion characteristics between the target and the imaging position are approaching motion, and the distance is getting closer. The change of the target motion trajectory is composed of azimuth, pitch and roll. That is, in the two-dimensional plane, the displacement relationship between the target in the current frame and the previous frame can be covered by the increments in the X and Y directions of the previous frame. Taking the 15 ms exposure period as the unit displacement, since the exposure period is at the ms level, the displacement of the target between two adjacent pixels is very small or even there may be no displacement. The target position (x, y) in the previous frame, and the target position in the current frame may cover four cases: (x, y), (x + 1, y), (x, y + 1), (x + 1, y + 1). Therefore, a 4-parameter motion model is designed. The compensation weight of the target gray-scale in the current frame after motion in the energy domain is obtained through global motion estimation to increase the gray-scale energy of the target area. Since the background and target gray-scales of the previous frame are added in the process of difference calculation in the time domain, it is necessary to filter out the energy of the previous frame in the energy domain after compensation and retain the target energy corresponding to the current frame.

[0096] The formula for motion compensation is expressed as follows:

[0097]

[0098]

[0099]

[0100] C k (i,j) = {f1 × GF t_k (i,j) + f2 × GF t_k (i - δ y ,j + δ z ) + f3 × GF t_k (i + δ y ,j - δ z ) + f4 × GF t_k (i - δ y ,j - δ z )}; (7)

[0101]

[0102] Among them, γ1 and γ2 are provided by the main control unit, which superimpose the grayscales of two adjacent frames of images before and after. γ1 performs target enhancement on the superimposed grayscale, and then obtains the grayscale difference GF t_k (i,j) of the two frames of images before and after by taking the difference. The grayscale difference distinguishes the background grayscale difference, the target grayscale difference, the grayscale difference between the target tail and the target, and the target-background grayscale difference. According to the target offset, the energy accumulation coefficients of three neighboring pixels after motion are estimated using the 4-parameter motion model of f. The coefficients are adjusted according to the division of the grayscale difference, and motion compensation of the target grayscale is achieved in the energy domain, weakening the influence of the time-domain characteristics on the target grayscale. The four energy accumulation coefficients will change in real time along with the motion characteristics of the target, such as height, speed, and maneuverability, and are calculated and bound in real time by the main control unit. δ y , δ z are the target offsets in the pitch and yaw directions relative to the previous frame after motion. Finally, γ2 is used to filter out the target grayscale energy of the previous frame image incorporated when calculating the grayscale difference before, and the grayscale energy after motion compensation of the current frame image is obtained.

[0103] α-β filtering is a dynamic system filtering method used to extract system state information from the output of sensors. It uses a first-order low-pass filter controlled by α and β coefficients to reduce the noise impact while retaining the system state information. In motion estimation, α-β filtering can be used to smooth the measured moving displacement data for more accurate prediction of the position in the next frame. The 4-parameter motion model is a model for the motion of an object on an image, typically used to predict the position of the object in the next frame. This model assumes that the motion of the object is uniformly accelerated in two directions (horizontal and vertical), and the velocity and acceleration are constant.

[0104] This model requires four parameters to describe the motion of the object: horizontal velocity, vertical velocity, horizontal acceleration, and vertical acceleration. Using this model, motion compensation can be achieved by predicting the position of the object in the next frame.

[0105] Combining these two methods, α-β filtering can be used to smooth the measured moving displacement data, and then these data are input into the 4-parameter motion model to predict the position of the object in the next frame. Finally, the motion compensation energy data is calculated based on the difference between the predicted position and the actual position. The α-β filtering coefficients are set to 0.04 and 0.3. The motion compensation energy data Ikm(n,:,:) calculated using α-β filtering and the 4-parameter motion model is the motion compensation energy data for each time step in n time steps. By increasing the gray value of the target area after motion, and through superimposing the pure gray background GE(1,n - 1) after gray level equalization, the preprocessed output image Gout(n,:,:) is obtained, which is the nth image after the preprocessing operation. Here, the first dimension n represents the number of images, the second dimension represents the number of rows of the image, and the third dimension represents the number of columns of the image.

[0106] Furthermore, 2 variables need to be read during the calculation process, namely the gray data after gray level enhancement of the previous frame image and the motion compensation energy data So, write RAM and read RAM operations involving these 2 variables are involved. Because parallel operations are required, 4 external SRAMs are needed, grouped in pairs of 2 for read-write ping-pong switching.

[0107] In addition, in the design of the image output algorithm, a conversion coefficient for the target gray level target gray energy needs to be added to convert the gray energy obtained by motion compensation into gray data. By superimposing the converted current frame target motion compensation image gray level on the equalized pure background floor of the previous frame, the gray data of the current frame image is obtained, that is, the final image. The algorithm is as shown in formula (9):

[0108]

[0109] In some feasible embodiments, it further includes:

[0110] S5. Perform image display:

[0111] The edge computing device processes, calculates, and stores the image data collected by the infrared camera in real time, and then returns the processing result to the cloud. The cloud analyzes the processing results of multi-machine collaboration and presents the analysis results on the mobile device according to user requirements. In addition, there is also a self-organizing network between edge computing devices to present the image results processed by edge devices on the mobile device.

[0112] S5.1. Data storage and transmission. The processed image data results can be temporarily or permanently stored inside the edge computing device. At the same time, the edge computing device transmits the processing results back to the cloud through network protocols.

[0113] S5.2. Cloud analysis and processing. After receiving the image data results transmitted back by the edge computing device, the cloud server uses more powerful computing capabilities, storage, and algorithms for further analysis and processing.

[0114] S5.3. Presentation of analysis results. The results processed by the cloud can be presented through the mobile device. Through real-time or offline reports, visual charts, alerts, or other forms of information presentation.

[0115] S5.4. Collaborative presentation between edge devices. The edge computing devices in the pod can also form a self-organizing network to share and present the image results processed by each other on the mobile device. This collaboration can improve real-time performance, accuracy, and reliability.

[0116] Through the above steps, the edge computing device of the pod and the cloud form a closely collaborative architecture, making the links of image data acquisition, real-time processing, cloud analysis, and mobile device presentation more efficient, reliable, and effective. This provides users with faster, more flexible, and intelligent image analysis and data processing services.

[0117] Combining all embodiments, the overall process and effect of the present invention are as follows: Output gray-level equalization is performed on the collected images, and the background and target gray levels of the entire image are equalized through the gray-level mean, weakening the contrast between the target signal and the background image; The improved median filter eliminates high-frequency noise points and retains the sharpness of the background and target edges; Motion compensation is achieved through the time-domain α-β filter and the 4-parameter motion model based on the median filter, estimating the energy compensation weight of the relative displacement of target pixels after image motion, and calculating the gray-level energy value of the pixels after motion.

[0118] To address the problems of complex image processing algorithms, large computational volume, and high energy consumption in the pod, the present invention employs the Jetson AGX Orin edge computing device. Jetson AGX Orin is the only development kit launched by NVIDIA in 2023. It is a GPU-accelerated computing platform with a computing power of about 200 TOPS and has the following advantages in the image processing of infrared images collected by the pod:

[0119] First, this device integrates a powerful GPU with excellent parallel computing capabilities, enabling it to process multiple image processing tasks simultaneously and accelerate the processing speed. Second, Jetson AGX Orin supports multiple mainstream programming languages such as C / C++ and Python and is compatible with a wide range of open-source libraries and tools, providing rich tools and resources for the development and implementation of image preprocessing algorithms. Third, it has rich software development tools and documentation support, providing a flexible software programming environment and architecture. Users can freely customize and optimize image preprocessing algorithms according to their needs to obtain better performance and results. Finally, this patent focuses on low power consumption and is suitable for the power-constrained pod environment. This enables it to reduce energy consumption while ensuring high performance. In summary, Jetson AGX Orin has outstanding advantages in the preprocessing of infrared images collected by the pod. Through its powerful parallel computing capabilities, rich image processing resources, and flexible customization capabilities, it can improve the clarity and utilization efficiency of pod infrared images.

[0120] During the process of infrared image acquisition by the optoelectronic pod, the target moves at a high speed during photography and the exposure time is relatively long. When the target rapidly undergoes temperature changes, or in cases of low light or poor lighting conditions, the target appears blurred or trailing. This patent provides a method based on motion compensation and enhancing the contrast between the moving target and the background. At the same time, based on motion compensation, this patent also proposes a method to equalize the target and the background to eliminate the trailing and blurring phenomena of the moving target.

[0121] As Figure 2 shown, an image processing system applicable to the edge computing device of the pod includes:

[0122] An acquisition module for acquiring images;

[0123] An equalization module for performing gray level equalization on the acquired images to obtain images after gray level equalization;

[0124] A filtering module for filtering the images after gray level equalization based on an adaptive filtering window size to obtain filtered images;

[0125] A compensation module for performing motion compensation on the filtered images to obtain the final images.

[0126] The content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented in the system embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0127] An image processing device applicable to pod edge computing devices:

[0128] At least one processor;

[0129] At least one memory for storing at least one program;

[0130] When the at least one program is executed by the at least one processor, the at least one processor implements an image processing method applicable to pod edge computing devices as described above.

[0131] The content in the above method embodiments is applicable to the device embodiments of the present invention. The functions specifically implemented in the device embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0132] A storage medium storing processor-executable instructions, where the processor-executable instructions, when executed by a processor, are used to implement an image processing method applicable to pod edge computing devices as described above.

[0133] The content in the above method embodiments is applicable to the storage medium embodiments of the present invention. The functions specifically implemented in the storage medium embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0134] The above is a specific description of the preferred embodiments of the present invention. However, the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of this application.

Claims

1. An image processing method, characterized in that, It includes the following steps: Collect an image; Perform gray level equalization on the collected image to obtain an image after gray level equalization; Based on an adaptive filtering window size, filter the image after gray level equalization to obtain a filtered image; Perform motion compensation on the filtered image to obtain a final image; The formula for gray level equalization is expressed as follows: Among them, G1(i,j) is the gray value of the pixel points in G k (i,j) in the image processing window that satisfy |G1(i,j) - GE last | ≤ G delta ; G delta is the binding parameter; G2(i,j) is the gray value of the pixel points in G k (i,j) in the image processing window that satisfy |G1(i,j) - G mean_last | > G delta ; N1 is the number of pixel points in G k (i,j) in the image processing window that satisfy |G1(i,j) - G mean_last | ≤ G delta ; N2 is the number of pixel points in G k (i,j) in the image processing window that satisfy |G1(i,j) - G mean_last | > G delta . i and j respectively represent the row index and column index of the pixels in the image. R end represents the ending row number of the processing window, and R start represents the starting row number of the processing window. C end represents the ending column number of the processing window, and C start represents the starting column number of the processing window; The step of filtering the image after gray level equalization to obtain a filtered image is specifically as follows: Calculate the difference between the gray level of the original pixel point in the image after gray level equalization and the gray level median; Determine the filtering window size according to the difference; Perform filtering based on the filtering window size to obtain a filtered image; The calculation formula for filtering is as follows: Where median(·) represents taking the median, U represents the full enumeration of data within a preset range around (i,j), and m and n represent the neighborhood indices relative to the current pixel point (i,j); The step of performing motion compensation on the filtered image to obtain a final image specifically includes: Based on the filtered image, calculate the gray level difference between two consecutive frames of images; Adjust the estimated energy accumulation coefficient according to the gray level difference to obtain an adjusted energy accumulation coefficient; Use the adjusted energy accumulation coefficient to filter the target gray level energy of the previous frame image to obtain compensated gray level energy; Convert the compensated gray level energy into image gray level data to obtain a final image; The formula for the motion compensation is expressed as follows: C k (i,j) = {f1 × GF t_k (i,j) + f2 × GF t_k (i - δ y , j + δ z ) + f3 × GF t_k (i + δ y , j - δ z ) + f4 × GF t_k (i - δ y , j - δ z )}; Among them, γ1 represents the first filtering coefficient, and γ2 represents the second filtering coefficient. represents the gray value of the previous frame image at position (i, j); GF t_k (i, j) represents the gray difference between two consecutive frames of images. represents the gray value of the current frame image at position (i, j); GF k (i, j) represents the gray value of the current frame after filtering at position (i, j); δ y represents the estimated value of the displacement of the image in the y direction; δ z represents the estimated value of the displacement of the image in the z direction; C k (i, j) represents the compensated gray energy. is I k (i, j) is the average of the absolute values, representing the motion compensation value of the k-th frame at position (i, j) after certain processing.

2. The image processing method according to claim 1, characterized in that It further includes: Perform preliminary recognition and regional division on the collected image to obtain image blocks of different regions.

3. The image processing method according to claim 2, wherein The step of performing preliminary recognition and regional division on the collected image to obtain image blocks of different regions specifically includes: Process the collected image based on the Canny edge detection algorithm and the target detection algorithm to obtain regional data and edge data; Based on the regional data and the edge data, calculate the feature point difference to obtain image blocks of different regions.

4. The image processing method according to claim 1, wherein It further includes: Implement parallel filtering through a multi-stage pipeline.

5. An image processing system, characterized in that, For executing an image processing method as described in claim 1, it includes: A collection module for collecting an image; An equalization module for performing gray level equalization on the collected image to obtain an image after gray level equalization; A filtering module for filtering the image after gray level equalization based on an adaptive filtering window size to obtain a filtered image; A compensation module for performing motion compensation on the filtered image to obtain a final image.

Images