REAL-TIME AND LOW-COST SATURATION AND BRIGHTNESS-BASED IMAGE ENHANCEMENT METHOD FOR NIGHTTIME CONDITIONS.

Abstract

The invention is an image processing method that enables the real-time and low-computational-cost enhancement of images obtained from RGB cameras for in-vehicle driver assistance systems in night and low-light conditions. Within the scope of the method, images from a standard RGB camera are converted to the HSV color space and separated into hue (H), saturation (S), and brightness (V) components. While the hue (H) component is kept constant to preserve the original colors, nonlinear, monotonically increasing transformations S'=S^(1 / 2) and V'=V^(3 / 4) are applied to the saturation component and brightness component respectively. This highlights areas of low saturation, balances the contrast between excessively bright and dark areas, and prevents the creation of artificial colors.

Description

REAL-TIME AND LOW TRANSACTION COST, EVEN AT NIGHT. SATURATION AND BRIGHTNESS-BASED IMAGE ENHANCEMENT METHOD Technical Area The invention aims to improve vehicle driving safety at night and in low light conditions. It relates to image processing and image enhancement methods. The invention specifically targets images obtained using standard RGB camera systems. real-time processing and adaptive color for improved visual perception. It is related to the improvement method. State of the Art There are differences in the current technology aimed at improving visibility in night driving conditions. Hardware and software-based solutions are used. Hardware-based systems among them infrared (IR) cameras, thermal imaging systems and active lighting. These include assisted night vision systems, but these systems are expensive. and has limited use due to integration challenges. Software-based In the field of image enhancement, histogram equalization and adaptive histogram equalization (CLAHE) are used. Gamma correction and Retinex-based methods are widely used. Additionally... In recent years, machine learning and deep learning-based low-light image enhancement Approaches have been developed. The current technical application number CN102231206A concerns automotive assisted driving. the brightness of night vision color images developed for use in systems It describes an image processing method aimed at enhancing quality. In the method, the input The image is first converted from the RGB color space to the YUV color space, then The brightness (Y) component of the S-curve is improved with the Retinex algorithm and selective linear It is enhanced by using a non-existent grayscale mapping method. The resulting The image is formed after the improved brightness components are combined using a weighted fusion method. The output is generated by converting it back to the RGB color space. This allows for night driving. 2 under these conditions, image details, brightness, and light-shadow information are preserved. The aim is to improve the driver's visual perception. In night driving conditions, both excessively bright and dim are present within the same scene. Effectively solving the problem of processing dark areas together is a complex issue. That's the problem. Because headlight beams propagate through space with directional guidance, there are areas outside the intended path. It can be completely dark. The headlights of oncoming vehicles are often visible. It creates excessive contrast, causing the driver's eye lens to shrink, resulting in vision loss. This can happen. Under these conditions, safe driving is the top priority. This priority... What needs to be done to meet this requirement is to apply the same principles used in conventional photographic images. Improving readability and distinguishability in the image instead of aesthetic approaches It should be. The processes to be performed should be approximately 25 frames per second, and each frame should be (HD). in real time in a camera system with approximately 2,000,000 pixels (in resolution) The fact that it is carried out in this way means that any desired method can be applied to this problem. It restricts. Histogram-based and Retinex-based methods are similar to the whole image. Because it applies transformations, local contrast imbalances can occur, and road signs, Critical driving information such as lane markings is not being adequately distinguished. Gamma correction Nonlinear methods, such as these, process all pixel values ​​using the same function. Because it transforms the scene, it fails to adequately account for on-stage differences. On the other hand, generating a dataset on which machine learning can be performed under night driving conditions. It's almost impossible. Capturing both the night and daytime states of the environment in a closed studio at that moment. It is possible to achieve this, but not under road driving conditions in such a studio environment. Revitalizing it would be a very costly approach. The model that would be obtained would also... Implementing this will also incur a transaction cost. In conclusion, due to the negative aspects described above and the current solutions, the subject matter... Due to its shortcomings, an improvement is needed in the relevant technical field. It has been observed. 3 Purpose of the Invention The invention is designed for use in in-vehicle driver assistance systems and has low potential. Low cost RGB camera-based systems provide low visibility during night driving conditions. It aims to improve visual quality. In this context, the invention relates to the HSV color space. image processing techniques based on processing of saturation and brightness components, and It is associated with low computational cost algorithms that run in real time. Furthermore, the invention enables real-time image processing that can run on embedded systems. solutions, applications such as in-vehicle display systems and night driving assistance systems It is designed for use in these fields. The invention is a low-cost and real-world solution that can operate through standard RGB camera systems. It offers a time-based image enhancement approach. Hardware-based night vision because it does not require additional sensors or special optical components compared to other systems. Applicable to a wide range of vehicle classes. Unlike global image processing methods used in the current state of the art, The proposed approach separates image components and is selective only on saturation. It performs an improvement. This results in an increase in color tone (Hue) and brightness (Value). The image is enhanced while preserving its natural structure. Furthermore, thanks to its nonlinear and monotonically increasing transformation structure, it is low Regions with low saturation are improved more strongly, while regions with high saturation are enhanced. It is protected in a controlled manner. This prevents the creation of artificial colors in the image and reduces perception. It prevents their illusions. In contrast to machine learning-based methods, the invention does not require a training dataset. It is capable of operating and is suitable for real-time applications with low transaction costs. It shows. The structural and characteristic features and all the advantages of the invention are described in detail below. This will make it clearer. 35 4 Detailed Description of the Invention In this detailed explanation, the preferred configurations of the invention are not merely for better understanding the subject. in order to facilitate understanding and without imposing any limiting effects It is explained. The invention allows images obtained with an RGB camera at night and in low-light conditions to be rendered in real life. an image that works on the HSV color space, enabling its improvement over time It is a processing method. The method separates the color components for each pixel, determining saturation and... It applies nonlinear transformations to the brightness, thereby altering the image. It improves its quality. Within the scope of this method, the obtained RGB image is first converted to HSV (Hue–Saturation–Value). This is converted into a color space. With this conversion, the hue (H), saturation (S), and color of each pixel are determined. The brightness (V) components are separated. This separation separates the color information from the brightness information. It allows for independent processing. In the basic processing step of the invention, the following is applied to the RGB components of each pixel: Transformations are applied: Normalization: r = R / 255 g = G / 255 b = B / 255 C = max(r, g, b) max C = min(r, g, b) min Δ = C - C max min Hue (H): If Δ = 0, then H = 0 If C = r H = 60 × ((g - b) / Δ mod 6) max If C = g then H = 60 × ((b - r) / Δ + 2) max 35 If Cmax = b H = 60 × ((r - g) / Δ + 4) Saturation (S): If C = 0 then S = 0 max otherwise S = Δ / Cmax Value (V): V = C max The Hue (H) component is preserved unchanged, while the saturation (S) component is nonlinear. The brightness (V) component is increased by transformation, recreated by a nonlinear transformation. It is scaled. In this context, saturation and brightness values ​​are calculated as follows: 1 / 2 Saturation value: S′=S 3 / 4 Brightness value: V′=V As a result of these transformations, low saturation values ​​are increased to a greater extent. Colors are made more vibrant while high saturation values ​​are controlled. It is preserved. Similarly, the transformation applied to the brightness component results in excessive brightness and By providing a more balanced distribution between excessively dark areas, the overall image quality... It increases its perceptibility. Thanks to the monotonically increasing nature of the transformation functions, the pixel values The relative order between them is preserved, and no artificial color creation or perceptual effects are present in the image. Distortion is prevented. Keeping the hue component constant ensures that the color tones remain unchanged. It ensures its protection. New RGB values ​​(R'G'B') for each pixel using processed HSV components. The calculations are made and the image is converted back to the RGB color space. The resulting image, It can be used directly through standard imaging systems. The method is suitable for parallel processing because it can be applied independently for each pixel. and supports real-time operation. Additionally, conversion processes are pre-calculated. It can be stored within a look-up table (LUT) structure. This allows 35. Computational costs are reduced, resulting in low processing load even in embedded systems. 6 It is feasible. The improved color values ​​will be read from memory using the M pointer. Assuming that access to the R'G'B' values ​​is calculated as follows: R'G'B' = M{ base address + offset value(RGB)} An RGB pixel contains 24 bits of information. Therefore, the LUT requires 16.77M x 3 Bytes = 50.33Bytes. A total of MBytes of memory is required. The base address is changed for different viewing modes. Switching to the LUT belonging to the mode can be done easily. The only cost of this switch is memory. Today, this level of memory represents a very low cost. In conclusion, the invention relates to the saturation and brightness components in the HSV color space. Color and image quality of nighttime images obtained through nonlinear transformations Real-time, low-cost image processing that improves brightness balance. It offers a method.

Claims

1. Images aimed at improving vehicle driving safety at night and in low light conditions. It is a processing and improvement method, and its characteristic is;  The red, green, and blue pixels of an image from an RGB camera by normalizing the values ​​of hue, saturation and brightness Separating the components by converting them to the HSV color space,  On the separated saturation component, regions with low saturation 1 / 2 to increase and protect high saturation regions S' = S application of mathematical transformation,  Extremely bright and dark regions on the separated brightness component. 3 / 4 In order to achieve balance between them, the mathematical transformation V' = V implementation,  Consistent color tone achieved through newly transformed saturation and brightness components. Calculating new R'G'B' values ​​for pixels using the component. It includes the steps of the process.

2. A method that complies with Claim 1, and its feature is that it reduces the conversion calculation cost. For the purpose of all conversions from the RGB color space to the R'G'B' color space It involves pre-calculating combinations and storing them in a lookup table, and The hardware cost of this process, excluding the RGB camera and processor, is 50.33 MByte. It is memory.