Image processing method and device based on crossing machine mode, equipment and medium

Abstract

This invention proposes an image processing method, apparatus, device, and medium based on a racing drone mode. Specifically, it includes: acquiring raw infrared image data based on the racing drone mode in real time, performing dual-layer spatial filtering denoising to generate an edge-preserving denoised image; performing high-intensity dynamic detail enhancement processing on the edge-preserving denoised image to generate a detail layer image; fusing the detail layer image with a grayscale-converted base layer image to generate an enhanced image; and performing block-based local contrast enhancement processing on the enhanced image to obtain a final high-resolution output image. This invention, through an innovative strategy combining dual-layer spatial filtering and block-based local contrast enhancement, significantly improves the clarity, contrast, and key target recognition of infrared images while ensuring real-time image processing.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an image processing method, apparatus, device and medium based on a racing drone mode. Background Technology

[0002] Infrared imaging technology is an imaging technique based on infrared radiation, capable of acquiring image information of objects even when visible light is insufficient or completely absent. Infrared imaging technology is widely used in military reconnaissance, night vision surveillance, medical diagnosis, environmental monitoring, and other fields. With technological advancements, the resolution and sensitivity of infrared imaging equipment are continuously improving, and its application scope is constantly expanding.

[0003] In recent years, FPV (First-Person View) drones have been increasingly used in recreational, competitive, and inspection applications. FPV drones typically employ first-person view (FPV) cameras, allowing operators to remotely control them via real-time video streaming. However, in the dynamic scenarios of high-speed flight, infrared image processing faces several challenges, including: how to improve the signal-to-noise ratio (SNR) of the image while meeting the requirements for real-time flight control response and efficient noise reduction algorithms; and how to present pilots with an intuitive thermal distribution image through dynamic range compression imaging while meeting high-temperature detection requirements.

[0004] Existing technologies typically employ a single filtering or enhancement algorithm, which struggles to effectively balance multiple requirements such as noise reduction, edge preservation, dynamic range compression, and local contrast enhancement while meeting real-time performance demands. This results in output images that are either noticeably noisy, have blurred details, or fail to highlight key targets, thus failing to meet the operational requirements of high-speed flight for racing drones. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes an image processing method, apparatus, device, and medium based on a racing drone model. Through an innovative strategy combining dual-layer spatial filtering and block-based local contrast enhancement, the clarity, contrast, and identification of key targets in infrared images are significantly improved while ensuring real-time image processing.

[0006] Firstly, this invention proposes an image processing method based on racing drone mode, which specifically includes the following steps: Step S1. Acquire raw infrared image data based on the racing drone mode in real time, and generate an edge-preserving and denoised image after performing dual-layer spatial filtering and denoising processing; Step S2. Perform high-intensity dynamic detail enhancement processing on the edge-preserving and denoising image to generate a detail layer image, and then fuse the detail layer image with the grayscale-converted base layer image to generate the enhanced image; Step S3. Perform block-based local contrast enhancement processing on the enhanced image to obtain the final high-resolution output image; The detail layer image is obtained by merging a first detail image obtained based on the difference method and a second detail image obtained based on the gradient method.

[0007] Furthermore, step S1 specifically includes: Step S11. Apply a multi-scale adaptive smoothing filter to the original infrared image data to perform the first layer of smoothing filtering; Step S12. Apply a bilateral filter to the image after the first layer of smoothing filtering and perform a second layer of edge-preserving filtering to generate an edge-preserving and denoised image; Specifically, a bilateral filter is used to perform a second layer of edge-preserving filtering to obtain the base layer image of the edge-preserving and denoising image.

[0008] Furthermore, step S11 specifically includes: using the multi-scale adaptive smoothing filter to perform strong smoothing processing on the uniform region in the original infrared image according to the gradient of the original infrared image, and weak smoothing processing on the edge region in the original infrared image, thereby removing strong noise in the original infrared image while retaining edge information.

[0009] Furthermore, in step S12, the bilateral filter performs a second layer of edge-preserving filtering within an N×N sliding window, where N is an odd number greater than 1. The spatial domain weights and value domain weights are obtained by looking up a table to calculate the filtered output value of the center pixel of the sliding window.

[0010] Furthermore, step S2 specifically includes: Step S21. Obtain the first detail image and the second detail image from the edge-preserving and denoising image using the direct difference method and the gradient method, respectively; Step S22. Merge the first detail image and the second detail image to generate a detail layer image; Step S23. After converting the base layer image of the edge-preserving and denoising image to grayscale using the histogram method, a base layer image to be fused is generated. The detail layer image and the base layer image to be fused are then fused to generate an enhanced image.

[0011] Furthermore, in the step of obtaining the first detail image by the difference method, Gaussian filtering and mean filtering are used to process the edge-preserving and denoising image to generate the first detail sub-image. After grayscale and normalization processing of the first detail sub-image, the first detail image is generated. In the step of obtaining the second detail image by the gradient method, an edge detection algorithm is used to process the edge-preserving and denoising image to generate the second detail sub-image. After grayscale and normalization processing of the second detail sub-image, gamma correction is then applied to generate the second detail image.

[0012] Furthermore, step S3 specifically includes: Step S31. Divide the enhanced image into multiple sub-blocks; Step S32. Perform histogram statistics independently for each sub-block and calculate the histogram clipping value based on the external input parameters; Step S33. Crop and redistribute pixels in the histogram of each sub-block, and calculate its cumulative histogram to generate the grayscale mapping table for that sub-block; Step S34. For any pixel in the enhanced image, calculate the difference in the grayscale mapping table using bilinear interpolation based on the pixel's coordinates to obtain the final high-resolution output image.

[0013] Secondly, the present invention also proposes an image processing apparatus based on a racing drone mode to perform the method as described in the first aspect, the image processing apparatus comprising: The image preprocessing module is used to acquire raw infrared image data based on the racing drone mode in real time, and generate edge-preserving and denoised images after performing dual-layer spatial filtering and denoising processing. The image fusion module is used to perform high-intensity dynamic detail enhancement processing on the edge-preserving and denoising image to generate a detail layer image, and then fuse the detail layer image with the grayscale-converted base layer image to generate an enhanced image. The image processing output module is used to perform block-based local contrast enhancement processing on the enhanced image to obtain the final high-recognition output image. The detail layer image is obtained by merging a first detail image obtained based on the difference method and a second detail image obtained based on the gradient method.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0015] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0016] The beneficial effects of this invention are: Compared with the prior art, the present invention has the following significant advantages: (1) Efficient noise reduction and detail preservation: The dual-layer spatial filtering mechanism works together to effectively suppress image noise and preserve edge and texture details to the greatest extent, solving the problem of "denoising is blurring" in traditional methods.

[0017] (2) High local contrast: The image block processing of the present invention specifically enhances the contrast of each local area of ​​the image, so that targets of different temperatures (regardless of size) can be clearly presented, greatly improving the target recognition efficiency of the pilot in complex backgrounds.

[0018] (3) Meets real-time requirements: The entire processing flow has been optimized through engineering (such as using lookup table method to implement bilateral filtering, frame-by-frame statistical histogram, etc.), and can be deployed on resource-constrained embedded platforms such as FPGA, meeting the stringent requirements of the racing drone for low latency image processing.

[0019] (4) Outstanding visual effects: The final output 8-bit infrared image has high definition, high contrast and distinct bright and dark details, providing the operator of the racing drone with intuitive and efficient visual information on thermal distribution. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the image processing method based on racing drone mode proposed in this invention; Figure 2 This is a schematic diagram of the image processing device framework based on the racing drone mode proposed in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, this embodiment proposes an image processing method based on racing drone mode, which specifically includes the following steps: Step S1. Acquire raw infrared image data based on the racing drone mode in real time, and generate an edge-preserving and denoised image after performing dual-layer spatial filtering denoising processing.

[0023] In this embodiment, step S1 specifically includes: Step S11. Apply a multi-scale adaptive smoothing filter to the original infrared image data to perform the first layer of smoothing filtering; Step S12. Apply a bilateral filter to the image after the first layer of smoothing filtering and perform a second layer of edge-preserving filtering to generate an edge-preserving and denoised image; Specifically, a bilateral filter is used to perform a second layer of edge-preserving filtering to obtain the base layer image of the edge-preserving and denoising image.

[0024] In this embodiment, a multi-scale adaptive smoothing filter is used to perform the first layer of smoothing filtering on the original infrared image data. By performing strong smoothing in the uniform area of ​​the image and weak smoothing in the edge area, strong noise in the image is removed while edge information is preserved.

[0025] To more clearly illustrate the technical solution of this application, the acquired original infrared image is denoted as... Furthermore, the image is 14 bits. During the first layer of smoothing filtering, the image gradient method is used to distinguish the smoothing intensity of details and noise in the image. The specific process is as follows: set up: Image gradient is represented as: Represents the original infrared image The pixel value located at the i-th row and j-th column. This represents the coordinate position. Assume the filtering function used for weak smoothing is... The strongly smoothing filter function is If T is the threshold, then the image after the first layer of smoothing filtering is represented as follows: : This represents the image after the first layer of smoothing filtering. The pixel value located at the i-th row and j-th column.

[0026] In practice, You can take a 5x5 size. Gaussian filter function, You can take a 5x5 size. The Gaussian filter function, where T can be 20.

[0027] Use a bilateral filter for the image A second layer of edge-preserving filtering is performed to generate an edge-preserving and denoised image. It should be noted that bilateral filtering is a non-linear filtering method that combines spatial proximity and pixel value similarity in an image, taking into account both pixel spatial differences and intensity differences to achieve edge-preserving noise reduction. In practical engineering, considering both logical resources and image quality, the bilateral filtering algorithm is performed within an N*N sliding window (N is an odd number greater than 1). A lookup table is used to obtain the spatial value kernel (Gaussian filter kernel) and the corresponding value of the value kernel at each point. The values ​​obtained from the lookup table are then used for the corresponding calculations to calculate the bilateral filtering pixel value at the center pixel of the sliding window.

[0028] In this embodiment, in order to achieve subsequent image fusion, after performing bilateral filtering on the image, the base layer image of the edge-preserving and noise-reducing image is obtained through the bilateral filter.

[0029] Step S2. Perform high-intensity dynamic detail enhancement processing on the edge-preserving and denoising image to generate a detail layer image, and then fuse the detail layer image with the grayscale-converted base layer image to generate the enhanced image.

[0030] In this embodiment, step S2 specifically includes: Step S21. Obtain the first detail image and the second detail image from the edge-preserving and denoising image using the direct difference method and the gradient method, respectively; Step S22. Merge and fuse the first detail image and the second detail image to generate a detail layer image; Step S23. After converting the base layer image of the edge-preserving and denoising image to grayscale using the histogram method, a base layer image to be fused is generated. The detail layer image and the base layer image to be fused are then fused to generate an enhanced image.

[0031] It should be noted that various methods can be used to obtain image details in practice. In this embodiment, for the rapid motion blur and thermal inertia boundary blur of the racing drone image, a single detail enhancement method is prone to amplifying noise or producing artifacts. This embodiment adopts a multi-path parallel extraction and intelligent fusion strategy to obtain image details. Combining the above content, the specific process is as follows: The first detail sub-image is obtained by the direct difference method. And expressed as: here and These are the 3*3 Gaussian filter operator and the 3*3 mean filter operator, respectively.

[0032] For the first detail sub-image Grayscale limiting and normalization are performed to obtain the first detail image. .

[0033] Gradient-based methods for obtaining second detail sub-images And expressed as: here and These are the Sobel operators in the x-direction and y-direction of the edge detection algorithm, respectively.

[0034] For the second detail sub-image After grayscale limitation and normalization, gamma correction is performed. Obtain the second detailed image Gamma correction here can enhance low and medium gradient values, thus enhancing blurred boundaries, and suppress high gradient values, thus suppressing sharp boundaries and noise.

[0035] For the first detail image Second detail image Merge the images to generate detail layer images. And expressed as: Here, the range of R is... .

[0036] The above steps achieve the processing of the detail layer image. To achieve image fusion, the acquired base layer image is processed using the AGC algorithm to convert the image from 16 bits to 8 bits of grayscale, thus obtaining the base layer image to be fused. In this embodiment, the AGC algorithm generally uses the histogram method for conversion, but this embodiment does not impose specific limitations.

[0037] To meet the image requirements of high-speed flight scenarios for racing drones, a DDE (Digital Detail Enhancement) approach is employed. This method fuses the detail layer image and the base layer image to be fused, resulting in an enhanced image. It should be noted that DDE, in image processing (especially in infrared imaging, medical imaging, and industrial vision), typically refers to Digital Detail Enhancement, a non-linear enhancement technique used to improve local contrast and texture details in images.

[0038] Step S3. Perform block-based local contrast enhancement processing on the enhanced image to obtain the final high-resolution output image.

[0039] In this embodiment, step S3 specifically includes: Step S31. Divide the enhanced image into multiple sub-blocks; Step S32. Perform histogram statistics independently for each sub-block and calculate the histogram clipping value based on the external input parameters; Step S33. Crop and redistribute pixels in the histogram of each sub-block, and calculate its cumulative histogram to generate the grayscale mapping table for that sub-block; Step S34. For any pixel in the enhanced image, calculate the difference in the grayscale mapping table using bilinear interpolation based on the pixel's coordinates to obtain the final high-resolution output image.

[0040] In this embodiment, the enhanced image is divided into M×N sub-blocks; the high-direction resolution of the detector used for image processing is known to be... Width resolution is Then the height of each sub-block is , width is .

[0041] Histograms are generated for each sub-block. , When counting the data in the m-th row and n-th column of the current statistics block, .

[0042] Count the total number of non-zero gray levels And adjust the generated histogram.

[0043] Through parameters Adjust the histogram clipping value and calculate the histogram clipping value using the total number of non-zero gray levels (count). : In this embodiment, This is an adjustable parameter, which can be input externally and its value can be determined based on practical experience. .

[0044] Use histogram clipping values Limit the histogram and count the total number of histogram pixels that exceed the clipping value. : The restricted histogram is And the histogram pixels exceeding the clipping value are allocated to the histogram according to a certain ratio. Then the number of pixels allocated to each non-zero gray level is: The adjusted histogram is as follows: Calculate the cumulative histogram And generate a mapping table .

[0045] Based on the pixel position, linear interpolation is used to process the M×N mapping tables. Interpolation is performed to obtain the final histogram mapping result. This results in a final, highly recognizable output image.

[0046] The following describes the solution of this embodiment in conjunction with the specific steps described above and the actual application of FPGA. The FPGA solution used is the Ti60F100 ultra-small 60K FPGA. The Ti60F100 has 60K logic cells, high-speed I / O that can be configured for various protocols, and built-in SPI Flash and Hyper-RAM, all contained in a tiny 5.5*5.5mm package with only a 0.5mm pitch.

[0047] The infrared detector selected is model VL0612VC. This detector is based on ceramic packaging technology, with a pixel size of 640x512 and a pixel pitch of 12μm. During FPGA engineering, the main focus was on the interaction and engineering of two important algorithm modules: first, the engineering of the dual-layer spatial filtering algorithm; and second, the implementation of the 8-bit image block algorithm based on the racing drone.

[0048] I. Engineering Application of Two-Layer Spatial Filtering Algorithm The first-layer spatial filtering is a Gaussian-based spatial filtering algorithm that can eliminate prominent noise in the image, achieving a preliminary smoothing effect. The second-layer spatial filtering is a bilateral filtering algorithm that can further suppress noise while maintaining the sharpness of image edges. The two spatial filtering layers work together to achieve good noise suppression results.

[0049] (1) First-layer spatial filtering The image data is windowed to form a 5x5 pixel matrix X1: Using the center pixel value (data22) of the matrix as the reference, and with w1 and w2 as coefficients respectively, the center point data is calculated. The gradient value was calculated using data from a central 3x3 area. and set a threshold =parameter1 (where parameter1 is an adjustable parameter, with an adjustment range of 0~31), the pixel values ​​in matrix X1 are... The coefficient coefxx for each pixel is set to 1, otherwise to 0. The following calculation is performed for each pixel: The calculated image data is windowed to form a 5x5 pixel matrix x 2: Calculate matrix If the standard deviation stdX2 is used as the benchmark for threshold setting, then the threshold... `stdX2` sets the threshold coefficient `coefxx2` for each pixel. Pixels in the matrix whose values ​​fall within the range of `data22±threshold2*2` have a threshold coefficient of 1; otherwise, they have a threshold coefficient of 0. The calculation is performed for each pixel. We obtain matrix X3: Based on matrices X2 and X3, obtain matrices X4 and X5: Calculate the mean values ​​of matrices X4 and X5, which are Avg_X4 and Avg_X5, respectively.

[0050] The basic local feature metric SD1 is calculated as follows: Where 0 ≤ SD1 ≤ 12383, then the value of the detail enhancement gain SD2 is determined based on the value of SD1: Calculate three numbers: noise suppression strength Edge protection weight and dynamic range compression threshold , respectively represented as: Here, parameter2 is an adjustable parameter with an adjustment range of 0~300, in order to achieve a better filtering and smoothing effect.

[0051] Filtered data is obtained from SD3, SD4, and SD5. : That is, it serves as the input for the second-layer spatial filter.

[0052] (2) Second-layer spatial filtering The second-layer spatial filtering is a bilateral filtering algorithm performed within a 3*3 sliding window. It uses a lookup table method to obtain the spatial range kernel (Gaussian filter kernel) and the range kernel template, thereby realizing the calculation of the bilateral filtering pixel value of the center pixel of the sliding window.

[0053] For a fixed spatial domain weight value Its spatial domain core The Gaussian filter kernel is fixed; the table below shows the Gaussian filter kernels used in the algorithm. ): II. Implementation of 8-bit Image Blocking Algorithm Based on FPV Camera The current DDE algorithm results in low image contrast. To improve image contrast, an 8-bit image block AGC algorithm is added after the existing DDE algorithm to enhance the contrast of the displayed image and increase the brightness and darkness levels. Specific methods used in the engineering process include: Create a dual-port RAM to store the grayscale histograms of the blocks.

[0054] Statistical histograms (histograms) of images are generated on a block-by-block basis.

[0055] Generate effective row signals for statistical image histograms on a block-by-block basis.

[0056] Count the frames: count the lower half of the image for odd-numbered frames and the upper half for even-numbered frames. Perform histogram statistics. The histogram height is 16 bits, limited to 65535.

[0057] Histogram Adjustment: Input parameters from the menu and calculate the histogram clipping value. At frame intervals, count the total number of histogram pixels exceeding the clipping value in the state machine. In state machines S5 and S6, distribute TotalExc evenly across all gray levels. In S6, calculate the total number of valid pixels, Histvnum.

[0058] Calculate the cumulative histogram and generate a grayscale mapping table: In state machine S7, calculate the cumulative histogram and generate grayscale data. Simultaneously, generate the write address for the grayscale mapping table. Create two dual-port RAMs to store the grayscale mapping tables for the upper and lower blocks respectively. In the state machine, write the calculated grayscale data into the corresponding block's grayscale mapping table.

[0059] Based on the pixel coordinates, bilinear interpolation is performed using the corresponding two histogram mapping tables to obtain the final histogram mapping result: When the row of the next frame is valid, the image data is used as the read address to read the data from the two grayscale mapping tables in RAM, and Table0 and Table1 are obtained after limiting the range. Based on the row and column counts, the distance from the pixel to the block center is calculated, and the bilinear weighting coefficients cofy1 and cofy2 are calculated based on the distance. Based on the row and column counts, it is determined which histograms vx0 and vx1 to use. Interpolation yields the final mapping result DATA1_O_mid1.

[0060] Add a switch for data output: Record the unprocessed data using the 8-bit block algorithm and synchronize the timing of the processed data. Switch the output data according to the algorithm switch.

[0061] Figure 2 This is a schematic diagram of an image processing device framework based on a racing drone mode according to an embodiment. The image processing device 200 includes: The image preprocessing module 201 is used to acquire raw infrared image data based on the racing drone mode in real time, and generate an edge-preserving and denoised image after performing dual-layer spatial filtering and denoising processing. Image fusion module 202 is used to perform high-intensity dynamic detail enhancement processing on the edge-preserving and denoising image to generate a detail layer image, and then fuse the detail layer image with the grayscale-converted base layer image to generate an enhanced image; The image processing output module 203 is used to perform block-based local contrast enhancement processing on the enhanced image to obtain the final high-recognition output image. The detail layer image is obtained by merging a first detail image obtained based on the difference method and a second detail image obtained based on the gradient method.

[0062] This embodiment also discloses an electronic device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of an image processing method based on a racing drone mode.

[0063] Specifically, the electronic device includes a processor and a memory connected via a system bus. The processor provides computing and control capabilities to support the operation of the entire electronic device. The memory may include non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. These computer programs can be executed by the processor to implement the image processing methods based on the racing drone mode provided in the following embodiments. The internal memory provides a cached runtime environment for the operating system computer programs in the non-volatile storage media.

[0064] The various modules in the apparatus provided in this application embodiment can be implemented in the form of a computer program. This computer program can run on a terminal or server. The program modules constituted by this computer program can be stored in the memory of the terminal or server. When the computer program is executed by a processor, it implements the steps of the method described in the embodiments of this application.

[0065] This embodiment also provides a computer-readable storage medium. One or more non-volatile computer-readable storage media containing computer-executable instructions, which, when executed by one or more processors, cause the processors to perform steps of an image processing method based on a drone-based mode. A computer program product containing instructions, which, when run on a computer, causes the computer to perform image processing based on a drone-based mode.

[0066] Any references to memory, storage, database, or other media used in the embodiments of this application may include non-volatile and / or volatile memory. Suitable non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which is used as external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. An image processing method based on racing drone mode, characterized in that, Specifically, the steps include the following: Step S1. Acquire raw infrared image data based on the racing drone mode in real time, and generate an edge-preserving denoised image after performing dual-layer spatial filtering denoising processing; Step S2. Perform high-intensity dynamic detail enhancement processing on the edge-preserving denoised image to generate a detail layer image, and fuse the detail layer image with the grayscale-converted base layer image to generate an enhanced image; Step S3. Perform block-based local contrast enhancement processing on the enhanced image to obtain the final high-recognition output image; The detail layer image is obtained by merging a first detail image obtained based on the difference method and a second detail image obtained based on the gradient method.

2. The image processing method based on racing drone mode according to claim 1, characterized in that, Step S1 specifically includes: Step S11. Applying a multi-scale adaptive smoothing filter to the original infrared image data to perform a first-layer smoothing filter; Step S12. Applying a bilateral filter to the image after the first-layer smoothing filter to perform a second-layer edge-preserving filter to generate an edge-preserving denoised image; Specifically, a bilateral filter is used to perform a second layer of edge-preserving filtering to obtain the base layer image of the edge-preserving and denoising image.

3. The image processing method based on racing drone mode according to claim 2, characterized in that, Step S11 specifically includes: using the multi-scale adaptive smoothing filter to perform strong smoothing processing on the uniform region in the original infrared image according to the gradient of the original infrared image, and weak smoothing processing on the edge region in the original infrared image, thereby removing strong noise in the original infrared image while retaining edge information.

4. The image processing method based on racing drone mode according to claim 2, characterized in that, In step S12, the bilateral filter performs a second layer of edge-preserving filtering within an N×N sliding window, where N is an odd number greater than 1. The spatial domain weights and value domain weights are obtained by looking up a table to calculate the filtered output value of the center pixel of the sliding window.

5. The image processing method based on racing drone mode according to claim 3 or 4, characterized in that, Step S2 specifically includes: Step S21. Obtain the first detail image and the second detail image from the edge-preserving and denoising image using the direct difference method and the gradient method, respectively; Step S22. Merge the first detail image and the second detail image to generate a detail layer image; Step S23. After converting the base layer image of the edge-preserving and denoising image to grayscale using the histogram method, a base layer image to be fused is generated. The detail layer image and the base layer image to be fused are then fused to generate an enhanced image.

6. The image processing method based on racing drone mode according to claim 5, characterized in that, In the step of obtaining the first detail image by the difference method, Gaussian filtering and mean filtering are used to process the edge-preserving and denoising image to generate the first detail sub-image. After grayscale and normalization processing of the first detail sub-image, the first detail image is generated. In the step of obtaining the second detail image by the gradient method, an edge detection algorithm is used to process the edge-preserving and denoising image to generate the second detail sub-image. After grayscale and normalization processing of the second detail sub-image, gamma correction is then applied to generate the second detail image.

7. The image processing method based on racing drone mode according to claim 6, characterized in that, Step S3 specifically includes: Step S31. Divide the enhanced image into multiple sub-blocks; Step S32. Perform histogram statistics independently on each sub-block and calculate the histogram clipping value based on external input parameters; Step S33. Crop and redistribute pixels in the histogram of each sub-block and calculate its cumulative histogram to generate the grayscale mapping table for that sub-block; Step S34. For any pixel in the enhanced image, calculate the difference in the grayscale mapping table based on the pixel's coordinates using bilinear interpolation to obtain the final high-resolution output image.

8. An image processing apparatus based on a racing drone mode, used to perform the method as described in any one of claims 1 to 7, characterized in that, The image processing device includes: The image preprocessing module is used to acquire raw infrared image data based on the racing drone mode in real time, and generate an edge-preserving and denoised image after performing dual-layer spatial filtering and denoising processing; the image fusion module is used to perform high-intensity dynamic detail enhancement processing on the edge-preserving and denoised image to generate a detail layer image, and fuse the detail layer image with the grayscale-converted base layer image to generate an enhanced image. The image processing output module is used to perform block-based local contrast enhancement processing on the enhanced image to obtain the final high-recognition output image. The detail layer image is obtained by merging a first detail image obtained based on the difference method and a second detail image obtained based on the gradient method.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images