FPGA-based vehicle-mounted image enhancement method

Abstract

The application relates to a kind of FPGA-based vehicle-mounted image enhancement methods, belong to video signal processing field.The application obtains the similar reverse to be enhanced image of foggy image by pixel inversion to the image to be enhanced, and does standardization processing, obtains the standardization reverse to be enhanced image;The average saturation of standardization reverse to be enhanced image is solved;According to the average saturation of standardization reverse to be enhanced image solved in step 2, adaptive correction parameter is determined;Adaptive correction parameter is brought into the saturation calculation of standardization reverse enhancement image, adjusts according to the brightness of overall image and the brightness of pixel, calculates transmissivity, and obtains enhancement image according to the defogging model framework.The application has good adaptability in extremely low light source, backlight, overall contrast is low but local light and dark contrast is obvious night scene and other complex vehicle-mounted scenes;Prior training and a large number of complex calculation processes are not needed for deep learning.

Description

Technical Field

[0001] This invention relates to the field of video signal processing technology, specifically to an FPGA-based in-vehicle image enhancement method. Background Technology

[0002] In vehicle-mounted video surveillance systems, PAL (Phase Alternating Amplitude) analog video sources, as a mature and widely used video signal standard, offer significant advantages and broad applicability. The PAL standard boasts high compatibility and stability, providing reliable video signal transmission in various complex environments. The hardware cost of PAL analog video sources is relatively low, and they are easy to integrate into existing vehicle systems, leading to their widespread adoption in the vehicle monitoring field. However, the lighting conditions in vehicle environments are complex and variable, especially in backlit or low-light conditions, where PAL video signals often experience image quality degradation.

[0003] Existing low-light image enhancement algorithms are mainly divided into frequency domain methods and spatial domain methods. Frequency domain methods transform the image from the spatial domain to the frequency domain, performing enhancement, filtering, and restoration operations on the image's spectrum in the frequency domain, and then returning to the spatial domain through an inverse transformation. However, due to its high computational complexity, large resource requirements, and poor real-time performance, it is often unsuitable for direct deployment on resource-constrained automotive embedded platforms. Spatial domain enhancement methods are a class of enhancement methods that directly manipulate image pixels and occupy an important position in image processing and visual perception enhancement. These include histogram processing, spatial filtering, and Retinex enhancement (a computational theory of color constancy perception). Among these, Retinex enhancement is widely used in low-light image enhancement because it can effectively simulate the human eye's perception mechanism of light and color, effectively improving image brightness, contrast, and detail preservation. Image enhancement methods based on the dehazing framework share some similarities with the Retinex enhancement method in their computational processing. In the literature CN202510292774.4, the dehazing framework is mainly used to achieve two-step estimation of atmospheric illumination intensity through a combination of decreasing cutting and genetic algorithm. Dark color prior and guided filtering are used to calculate transmittance. Finally, gamma correction is used for adaptive compensation of brightness and contrast.

[0004] This method has certain advantages in image detail restoration, but it does not fully consider the hardware implementation in practical engineering, and the process is relatively complex and difficult to meet real-time requirements. In the paper CN201811220177.7, the main feature of the method is based on a dehazing framework and uses a neural network to predict transmittance. Specific steps such as feature extraction using a CNN model, local extremum block processing, and dual-guided thinning have high requirements for computational resources and real-time performance. If used in embedded devices or edge computing scenarios, it suffers from low operating efficiency. Furthermore, the high computational complexity and large memory consumption of such methods during the inference phase make them unsuitable for direct deployment in resource- and power-constrained automotive image embedded platforms or real-time systems.

[0005] Based on this, the present invention is proposed. Summary of the Invention

[0006] Dehazing models have been applied in the field of image enhancement. Their core principle is to restore degraded images based on atmospheric scattering models. The images to be enhanced directly acquired by the vehicle system under low light conditions are highly similar to foggy images in terms of visual performance and degradation mechanism. Both exhibit low overall contrast and blurred or unrecognizable details.

[0007] Based on the above, this invention proposes an FPGA-based vehicle image enhancement method. By inverting the image to be enhanced, dark areas are transformed into bright areas, giving it degradation characteristics similar to foggy images. By adjusting atmospheric light intensity and transmittance parameters, real-time adaptive enhancement of backlit and low-light scenes is achieved based on image saturation.

[0008] An FPGA-based in-vehicle image enhancement method includes the following steps:

[0009] Step 1: Invert the pixels of the image to be enhanced to obtain an inverted image similar to the foggy image, and then normalize it to obtain a normalized inverted image to be enhanced.

[0010] Step 2: Calculate the average saturation of the standardized inverted image to be enhanced;

[0011] Step 3: Determine the adaptive correction parameters based on the average saturation of the standardized inverted image to be enhanced obtained in Step 2;

[0012] Step 4: Input the adaptive correction parameters into the saturation calculation of the normalized inversion enhanced image and adjust them according to the overall image brightness and pixel brightness;

[0013] Step 5: Substitute the inverted image to be enhanced from Step 1, the average saturation of the normalized inverted image to be enhanced from Step 2, and the saturation of the normalized inverted enhanced image from Step 4 into the calculation to obtain the transmittance.

[0014] Step 6: Based on the inverted image to be enhanced from Step 1 and the transmittance obtained in Step 5, input them into the defogging model framework to obtain the enhanced image.

[0015] As a further aspect of the present invention, in step 1, the analog-to-digital conversion of the PAL vehicle image input is completed by the decoder. The YCbCr 4:2:2 image format data output by the decoder is sequentially restored and converted to YCbCr 4:4:4 and RGB8:8:8 by the FPGA to obtain the image to be enhanced, which serves as the image input source.

[0016] As a further aspect of the present invention, in step 1, the image to be enhanced is pixel-inverted, and the inverted image to be enhanced is fed into the dehazing framework for processing, as shown in the following expression:

[0017] = ;

[0018] = ;

[0019] = ;

[0020] Where x represents the pixel position in the image coordinate system, Indicates the image to be enhanced. This represents the three color channels: RGB. Indicates an enhanced image. Indicates by Estimated atmospheric light intensity, This indicates that the image to be enhanced has been inverted. Transmittance represents the light's transmission rate after it has been reflected from an object and reached the camera. This represents the inverted image after enhancement processing.

[0021] As a further aspect of the present invention, in step 1, the expression for the standardization process is as follows:

[0022] ;

[0023] , ;

[0024] in, This indicates a standardized inverted image to be enhanced. This represents a normalized inverted enhanced image.

[0025] As a further aspect of the present invention, in step 2, saturation is defined as:

[0026] ;

[0027] ;

[0028] Where x represents the pixel position in the image coordinate system, This represents the minimum value among the R, G, and B color component channels of the input image. This represents the maximum value in the R, G, and B color component channels of the input image;

[0029] Indicates average image saturation. This represents the total number of pixels in each frame of a PAL video.

[0030] As a further aspect of the present invention, in step 3, the adaptive correction parameter is... satisfy:

[0031] =

[0032] Normalize and invert the average saturation of the image to be enhanced Images with a value greater than or equal to 0.5 are classified as backlit images, and the average saturation of the image to be enhanced is normalized and inverted. Images with a value less than 0.5 are classified as low-light images; settings Adjusting the normalized inversion under backlight conditions enhances image saturation. Adjusting the normalized inversion to enhance image saturation under low light conditions;

[0033] Adaptive correction parameters are used when processing backlight images. The expression is as follows:

[0034] = ;

[0035] In the formula, This indicates that the saturation of the image to be enhanced has been normalized and inverted. This represents the average saturation of the image to be enhanced, which is then normalized and inverted. Indicates the dark segment adjustment factor. Indicates the adjustment factor for the bright segment. Represents the dark area offset constant. Let p represent the bright area offset constant, p represent the dark area nonlinear response index, and q represent the bright area enhancement suppression index; preferably, 0.1≤p≤0.9 and 1.5≤q≤3.

[0036] The adaptive correction parameter set for the backlight image is set to a value of [value to be filled in]. , , , ;

[0037] =

[0038] , , , ,

[0039] ;

[0040] This indicates an enhanced intermediate reference value. , This indicates the enhancement lower limit factor. Indicates the enhancement upper limit factor;

[0041] The adaptive correction parameter set for low-light images is set to a value of [value missing]. , , , ;

[0042] =

[0043] , , , ,

[0044] .

[0045] As a further aspect of the present invention, in step 4, the expression for obtaining the saturation of the standardized inversion enhanced image is set as follows:

[0046]

[0047] =

[0048] This represents normalized inversion to enhance image saturation. The target saturation after enhancing the saturation of the standardized inverted image in the piecewise expression is abbreviated as: , Normalized inversion of image saturation to be enhanced and target saturation constitute. It controls the intensity of the enhancement;

[0049] As a further aspect of the present invention, in step 5, the transmittance The expression to be evaluated is:

[0050] .

[0051] As a further aspect of the present invention, in step 6, Set to a fixed value [1,1,1] to enhance the image. The expression to be evaluated is:

[0052] .

[0053] The present invention has the following benefits and advantages:

[0054] 1. Image enhancement based on dehazing models shares some similarities with Retinex theory methods. While traditional Retinex methods can improve the brightness of dark areas in images, they are prone to halo effects and grayscale distortion in extremely dark scenes. Improved multi-scale or adaptive Retinex methods can alleviate these problems to some extent, but there is still room for improvement in the enhancement effect in extremely low light source areas. This invention can accurately and adaptively adjust the transmittance of different areas based on brightness and saturation, thereby avoiding halo effects and grayscale distortion. It has good adaptability in complex automotive scenes such as extremely low light source, backlight, and night scenes with low overall contrast but significant local brightness and darkness contrast.

[0055] 2. While deep learning has made significant progress in low-light image enhancement, most methods rely on large amounts of training data, have complex network structures, and require high computational resources, making it difficult to meet the real-time and deployment requirements of automotive embedded platforms. Furthermore, their generalization ability across different scenarios is limited. To address these issues, this invention improves automotive image quality while fully considering the feasibility and engineering feasibility of implementation on FPGA (Programmable Logic Device) embedded hardware platforms. The method of this invention is based on a dehazing model and an adaptive saturation adjustment mechanism, eliminating the need for prior training and extensive complex calculations in deep learning, thus possessing greater versatility, stability, and platform adaptability. To meet hardware acceleration requirements, a modular and pipelined design is adopted. Each processing stage, such as transmittance estimation, saturation adjustment, and enhanced image restoration, is executed sequentially in a time-driven manner, resulting in low pixel-level latency throughout the image processing. Enhancement processing can be completed in just a few pixel clock cycles, fully meeting the response requirements of real-time application scenarios. In addition, the nonlinear calculations involved in this invention are all efficiently implemented using lookup tables, reducing computational complexity and improving the feasibility and stability of engineering implementation. Meanwhile, during the FPGA implementation, the system is very resource-friendly. On a medium-sized FPGA platform represented by the XCAU15P-UBVA368 (a device in the Artix UltraScale+ FPGA series launched by Xilinx), the overall usage of logic units such as lookup tables and flip-flops, as well as internal storage resources, remains at a low level, not exceeding 1 / 10 of the total chip resources. This fully demonstrates that this method has advantages such as low resource overhead, controllable power consumption, and simple deployment. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the workflow of an FPGA-based in-vehicle image enhancement method.

[0057] Figure 2 To handle the adaptive correction parameter function relationship for backlit images;

[0058] Figure 3 The adaptive correction parameter function relationship for processing low-light images;

[0059] Figure 4 To standardize and invert the relationship of the image saturation function. Detailed Implementation

[0060] The present invention will be described in detail below with reference to specific embodiments. These embodiments are merely some, not all, implementations of the present invention. All other implementations obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] An FPGA-based in-vehicle image enhancement method, the workflow of which is as follows: Figure 1 As shown:

[0063] First, the GM7150BC decoder completes the analog-to-digital conversion of the 720*576 PAL vehicle image input, restoring the line and field signals and valid image data. The image format output by the GM7150BC decoder includes YCbCr 4:2:2 image format data.

[0064] The FPGA sequentially restores and converts the input YCbCr 4:2:2 image format data to YCbCr 4:4:4 and RGB 8:8:8 to obtain the image to be enhanced, which serves as the image input source. After a series of operations / processes, including image normalization and inversion, normalization and inversion of the image to be enhanced average saturation, adaptive parameter calculation, normalization and inversion of the enhanced image saturation, adaptive transmittance calculation, and enhancement image calculation, the enhanced image is obtained.

[0065] YCbCr 4:2:2 is a digital video color encoding format where luminance (Y) is sampled at full resolution, while chrominance (Cb / Cr) is sampled at half the horizontal direction. YCbCr 4:4:4 is a high-fidelity color difference format where both luminance and chrominance components are sampled once per pixel. RGB 8:8:8 is an uncompressed full-color format where each pixel uses 8 bits to represent the red (R), green (G), and blue (B) channels, for a total of 24 bits per pixel. The GM7150BC decoder supports analog video inputs such as PAL, converting them to digital video format and outputting YCbCr 4:2:2 format data.

[0066] Secondly, the normalization and inversion operation of the image to be enhanced directly complements each pixel of the input image to be enhanced, and normalizes the image by shifting it right by 8 bits. Then, adaptive parameters are calculated based on the average saturation of the normalized and inverted image to be enhanced obtained from the previous image. The saturation of the normalized and inverted enhanced image is adjusted by the adaptive parameter calculation to achieve adaptive enhancement of image pixels in backlight and low light source scenarios. Adaptive transmittance calculation is then performed, and the solved adaptive transmittance is used in the dehazing model to achieve image enhancement.

[0067] Finally, the BT656 timing standard is restored. After BT656 timing recovery, the image is output, and the enhanced vehicle-mounted image signal is provided to the SOC for processing and the monitoring terminal for display. BT656 is a parallel video transmission standard that defines the timing and signal format of YCbCr digital video (including embedded horizontal and vertical sync signals), commonly used for PAL standard-definition video transmission. SOC is a system-on-a-chip used for comprehensive processing in terminal devices.

[0068] Example 2

[0069] Step 1: Since the image to be enhanced and the hazy image are similar, this step involves inverting the image to be enhanced and feeding it into the dehazing framework for processing. The expression is as follows:

[0070] = ;

[0071] = ;

[0072] = ;

[0073] Where x represents the pixel position in the image coordinate system, Indicates the image to be enhanced. This represents the three color channels of RGB (an additive color model where R represents red, G represents green, and B represents blue, and various colors are generated by adjusting the intensity ratio of these three colors). Indicates an enhanced image. Indicates by Estimated atmospheric light intensity, This indicates that the image to be enhanced has been inverted. Transmittance represents the light's transmission rate after it has been reflected from an object and reached the camera. This represents the inverted image after enhancement processing.

[0074] Comparative research revealed that setting the atmospheric illumination intensity to a constant value [1,1,1] under low-light conditions, compared to theoretically calculating the atmospheric illumination intensity based on dark channel priors, produces the same image enhancement effect as observed by the human eye. Furthermore, the corresponding PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index) indices show no significant difference. In the FPGA (Field-Programmable Gate Array) hardware implementation, the theoretical method of calculating atmospheric illumination intensity based on dark channel priors inevitably incurs significant latency and consumes considerable logic resources. Therefore, in practical calculations, the atmospheric illumination intensity is set to a constant value. After obtaining the atmospheric illumination intensity, this step standardizes the above formula, as shown below:

[0075] ;

[0076] , ;

[0077] in, This indicates a standardized inverted image to be enhanced. This represents a normalized inverted enhanced image.

[0078] Step 2: In the image enhancement method of the dehazing frame, the transmittance affects the illumination enhancement effect. If the transmittance cannot be obtained based on the brightness of the entire image, it may lead to over-enhancement or under-enhancement. This invention adopts an adaptive transmittance calculation method based on saturation.

[0079] Saturation is used to represent color purity and is closely related to the brightness characteristics of an image. In low-light environments, insufficient light results in lower color purity, making the image appear dark and with low saturation. Under normal or high-light conditions, saturation is usually higher, making image areas brighter and more colorful. Therefore, the average saturation of the entire image can be used as an indicator of the overall brightness of the image. Saturation is defined as:

[0080] ;

[0081] ;

[0082] This step calculates saturation based on the above formula, where x represents the pixel position in the image coordinate system. This represents the minimum value among the R, G, and B color component channels of the input image. This represents the maximum value among the three color component channels R, G, and B of the input image.

[0083] Indicates average image saturation. This represents the total number of pixels in each frame of a PAL video. The hardware implementation outputs the average saturation of the image at the end of the current frame transmission through pipelined calculation.

[0084] Step 3: The PAL video in the vehicle system has interlaced scanning characteristics. For each frame, the odd-field signal has valid data in lines 23-310, and the even-field signal has valid data in lines 336-623. The saturation difference between frames and fields in continuously acquired and transmitted video data is negligible. To ensure real-time processing and avoid buffered calculations, this step uses the average saturation of the previous field image to determine the category of the image to be enhanced. The image to be enhanced can be divided into backlit images and low-light images. Extensive experiments have shown that backlit images contain strongly lit backgrounds or high-contrast areas, resulting in significantly higher saturation. Low-light images, due to their near-grayscale colors, exhibit lower saturation. Therefore, average saturation can serve as an effective criterion for distinguishing lighting types. Based on statistical analysis, an empirical classification threshold of 0.5 is proposed, along with an adaptive correction parameter. The average saturation of the image to be enhanced will be normalized and inverted. Images with a value greater than or equal to 0.5 are classified as backlit images, and the average saturation of the image to be enhanced is normalized and inverted. Images with a value less than 0.5 are classified as low-light images. Adjusting the normalized inversion under backlight conditions enhances image saturation. The saturation of the image is enhanced by revising the normalized inversion under low light conditions, thereby affecting the extent of illumination enhancement of the image.

[0085] =

[0086] Adaptive correction parameters are used when processing backlight images. The expression is as follows:

[0087] = ;

[0088] In the formula, This indicates that the saturation of the image to be enhanced has been normalized and inverted. This represents the average saturation of the image to be enhanced, which is then normalized and inverted. Indicates the dark segment adjustment factor. Indicates the adjustment factor for the bright segment. Represents the dark area offset constant. This represents the bright area offset constant.

[0089] In this model, the unknown parameter p represents the nonlinear response exponent in dark areas, and q represents the enhancement suppression exponent in bright areas. If p > 1, the function grows too slowly in the low range, weakening the enhancement effect in dark areas; if p is too small, close to zero, the response will be too concentrated in extremely low saturation, resulting in a narrow enhancement target. Therefore, p needs to be balanced between enhancement coverage and selectivity between 0.1 and 0.9. If q < 1, the curve changes too slowly, and the enhancement area is not obvious; if q > 5, the curve changes abruptly, easily causing local enhancement jumps in bright areas. In actual measurements, 1.5 to 3 is a relatively balanced range.

[0090] Based on the above constraints, the construction relationship between the parameters is obtained as follows:

[0091] , , , ,

[0092] ;

[0093] This indicates that the intermediate reference value is enhanced to reflect the overall brightness trend of the image, therefore... As a supporting condition for piecewise functions, This indicates the enhancement lower limit factor. This represents the enhancement upper bound factor, satisfying the boundary conditions. , .

[0094] like Figure 2 As shown in the figure, the average saturation of the image to be enhanced is listed in the normalized inversion. Typical value cases. The values ​​are taken sequentially from 0.5 to 0.9, with intervals of 0.1. The larger, = The slower the descent, the more the average saturation of the image to be enhanced is normalized and inverted. A higher value indicates a brighter background light, resulting in lower brightness in darker areas of the image, requiring a higher value. To make adjustments, that is The larger, The slower the descent, the better. Conversely, a lower average saturation in the image to be enhanced indicates a darker background and brighter dark areas in the image, requiring a lower saturation level. To make adjustments, that is The smaller, The faster the decrease. A p value in the range of 0.6 to 0.8 has a good adaptation effect on backlit images. P=0.7 and q=2 were selected to ensure that the adaptive correction parameters are distinct but not abrupt. At the same time, to provide a safe and adaptable range of adaptive correction parameters, a setting was made... , .

[0095] Adaptive correction parameters used when processing low-light images The expression is as follows:

[0096] = ;

[0097] Due to the normalization and inversion of the average saturation of the image to be enhanced under low light conditions The brightness is relatively low. In order to flexibly adjust the position and width of the enhancement area according to the overall brightness of the image, the brightness is relatively low. As supporting conditions for piecewise functions, similarly, the construction relationship between parameters is obtained based on the constraints:

[0098] , , , ,

[0099] ;

[0100] like Figure 3 As shown in the figure, the average saturation of the image to be enhanced is listed by normalization and inversion. Typical value cases. The values ​​are taken sequentially from 0.1 to 0.4, with an interval of 0.1. The larger, The faster the decrease. When the normalization inversion of the low-light image is performed, the average saturation of the image to be enhanced is... A higher setting indicates that the image is brighter overall and requires a lower setting. To make adjustments, that is The larger, The faster the saturation decreases, the better. Conversely, a lower average saturation in the image to be enhanced indicates an overall darker image, requiring higher saturation levels. To make adjustments, that is The smaller, The slower the descent, the better. A p-value in the range of 0.1 to 0.3 shows good adaptability to low-light images; therefore, p=0.2 and q=2 were selected. , .

[0101] Step 4: Derive the defogging frame and take the maximum / minimum value to obtain the transmittance. Enhance image saturation with normalized inversion The relationship is as follows. This step enhances the brightness of the image by adjusting the saturation of the standard inversion enhanced image. The higher the saturation of the standard inversion enhanced image, the lower the transmittance, indicating a brighter enhanced image; conversely, the lower the saturation of the standard inversion enhanced image, the higher the transmittance, indicating a darker enhanced image.

[0102]

[0103] Among them, transmittance , This represents the maximum value among the R, G, and B color component channels of a normalized inverted enhanced image. This represents the maximum value of the R, G, and B color component channels in the image to be enhanced through normalization and inversion. Obviously... Less than or equal to Based on the above formula, the constraint condition between the saturation of the normalized inversion enhanced image and the saturation to be enhanced by normalization inversion is obtained:

[0104]

[0105] This step is based on the above constraints and substitutes the adaptive correction parameters obtained in step 3. The saturation of the normalized inversion enhanced image is obtained. The expression for the saturation of the normalized inversion enhanced image is as follows:

[0106]

[0107] =

[0108] To enhance the target saturation without weakening the original image's saturation, the enhancement process only stretches and does not compress. If the saturation is less than that of the image to be enhanced by normalization and inversion, the original value is retained directly. Some parts undergo nonlinear mapping processing, such as Figure 4 As shown, piecewise expressions enable rapid enhancement of dark areas in images and gentle, gradual enhancement of bright areas. The figure illustrates the normalized inversion of the average saturation of the image to be enhanced. Typical value cases, The values ​​are taken sequentially from 0.1 to 0.9, with an interval of 0.1. The larger the value, the more the normalized inversion enhances the image's saturation. Compared to normalizing and inverting the saturation of the image to be enhanced The more it increases, the more the expression will be in the left-hand interval. The upward-convex parabola is used to quickly enhance the visual effect in low-saturation areas. In the left-hand section... 1] The downward convex parabola is used to control the bright areas from being overexposed and dazzling, and the smooth transition avoids the enhancement effect being abrupt. It controls the intensity of the enhancement. The larger the area of ​​low saturation, the more obvious the enhancement effect.

[0109] Step 5: This step calculates the transmittance based on the saturation of the normalized inversion enhanced image obtained in Step 4. The transmittance is then calculated using the following expression. .

[0110]

[0111] Step 6: This step is based on the transmittance obtained in Step 5. By Set to a fixed value [1,1,1], and finally, set the image to be enhanced. Substitute the following expression to obtain the enhanced image. .

[0112] .

[0113] Furthermore, it should be understood that those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. An FPGA-based vehicle-mounted image enhancement method, characterized in that, The method comprises the following steps: Step 1, pixel inversion is performed on the to-be-enhanced image to obtain an inverted to-be-enhanced image similar to the foggy image and to perform standardization processing to obtain a standardized inverted to-be-enhanced image; Step 2, the average saturation of the standardized inverted to-be-enhanced image is calculated; Step 3, the adaptive correction parameter is determined according to the average saturation of the standardized inverted to-be-enhanced image calculated in step 2; Step 4, the adaptive correction parameter is brought into the saturation calculation of the standardized inverted to-be-enhanced image, and the brightness of the overall image and the brightness of the pixel are adjusted; Step 5, the transmittance is calculated according to the inverted to-be-enhanced image in step 1, the average saturation of the standardized inverted to-be-enhanced image in step 2, and the saturation of the standardized inverted to-be-enhanced image in step 4; Step 6, the enhanced image is obtained by bringing the inverted to-be-enhanced image in step 1 and the transmittance calculated in step 5 into the defogging model framework. 2.The FPGA-based vehicle-mounted image enhancement method of claim 1, wherein: In step 1, the analog-digital conversion of the PAL vehicle-mounted image input is completed by a decoder, and the YCbCr 4:2:2 image format data output by the decoder is sequentially restored and converted to YCbCr 4:4:4 and RGB 8:8:8 by an FPGA, so as to obtain the to-be-enhanced image. 3.The FPGA-based vehicle-mounted image enhancement method of claim 1, characterized in that: In step 1, the to-be-enhanced image is pixel-inverted, and the inverted to-be-enhanced image is brought into the defogging framework for processing, and the expression is as follows: = ； = ； = ； where x denotes a pixel position in an image coordinate system, denotes an image to be enhanced, denotes RGB three color channels, denotes an enhanced image, denotes an image to be enhanced by estimated atmospheric light intensity, denotes an inverted image to be enhanced, denotes a transmittance of light that propagates to a camera through reflection of an object, denotes an inverted image after enhancement processing.

4. The FPGA-based vehicle-mounted image enhancement method according to claim 3, characterized in that: In step 1, the expression of the standardization processing is as follows: ； ， ； wherein, denotes the normalized inverted image to be enhanced, denotes the normalized inverted enhanced image.

5. The FPGA-based vehicle-mounted image enhancement method according to claim 4, characterized in that: In step 2, the saturation is defined as: ； ； wherein x denotes a pixel position in an image coordinate system, denotes the minimum value in the input image R, G, B three color component channels, denotes the maximum value in the input image R, G, B three color component channels; represents the average image saturation, represents the total number of pixels per field image for PAL video.

6. The FPGA-based vehicle-mounted image enhancement method according to claim 5, characterized in that: In step 3, the adaptive correction parameter satisfies: = ； standardizing the average saturation of the inverse to-be-enhanced image images greater than or equal to 0.5 are classified as backlit images, and the average saturation of the inverse to-be-enhanced image is standardized images less than 0.5 are classified as low-light images; setting standardizing the saturation of the inverse to-be-enhanced image in the backlit case, standardizing the saturation of the inverse to-be-enhanced image in the low-light case; Adaptive correction parameters for processing backlit images The expression is as follows: = ； wherein denotes the normalized inverse saturation of the image to be enhanced, denotes the normalized average saturation of the image to be enhanced, denotes the dark segment adjustment factor, denotes the light segment adjustment factor, denotes the dark segment offset constant, denotes the light segment offset constant, p denotes the dark region non-linear response exponent, and q denotes the light region enhancement suppression exponent; The adaptive correction parameter set for the backlight image is valued and set as , , , ; = ； ， ， ， ， ； denotes an enhanced intermediate reference value, , denotes an enhanced lower bound factor, denotes an enhanced upper bound factor; The adaptive correction parameter set for the low-light image is valued and set as , , , ; = ； ， ， ， ， 。 7. The FPGA-based vehicle-mounted image enhancement method according to claim 6, characterized in that: In step 4, the expression for calculating the saturation of the standardized inverted to-be-enhanced image is set as: ； = ； representing the normalized inverse enhanced image saturation, is a target saturation for the normalized inverse enhanced image saturation in the piecewise expression, abbreviated as , is composed of the normalized inverse enhanced image saturation and the target saturation ; is a control of the intensity of the enhancement.

8. The FPGA-based vehicle-mounted image enhancement method according to claim 7, characterized in that: In step 5, the transmittance The expression is found to be: 。 9. The FPGA-based vehicle-mounted image enhancement method according to claim 8, characterized in that: In step 6, Set to constant value [1,1,1], enhance image The expression is found to be: 。

Images