CN-122023187-A - Illumination self-adaptive image preprocessing method and system for Shan Mugao-step sputtering SLAM

CN122023187ACN 122023187 ACN122023187 ACN 122023187ACN-122023187-A

Abstract

The invention discloses an illumination self-adaptive image preprocessing method and system for Shan Mugao S sputtering SLAM, and belongs to the technical field of computer vision and robots. Aiming at the problem that the front end of the existing SLAM system depends on a photometric consistency assumption, positioning drift and mapping artifacts are easy to generate under severe illumination fluctuation, the method decouples the processing flow into two dimensions, namely global dimension and local dimension. In CIELAB color space, smoothing correction of global exposure is realized through a continuous interpolation algorithm, a multi-factor joint constraint mechanism is introduced to adaptively calculate a dynamic threshold value of local contrast enhancement, and medium-frequency edge information is directionally promoted by combining a Gaussian differential sharpening technology. The method enhances the robustness of the system in the illumination abrupt change and extremely dark environment, remarkably improves the SLAM positioning precision, improves the image rendering capability when the 3D Gaussian sputtering is used for scene reconstruction, and effectively eliminates the suspended noise points and the geometric holes in the map.

Inventors

SUN XIAO
HONG WEIJIAN
ZHANG WANWEI
LI XINGXING

Assignees

武汉大学

Dates

Publication Date: 20260512
Application Date: 20260330

Claims (10)

1. An illumination adaptive image preprocessing method for Shan Mugao s sputtering SLAM, characterized by comprising: Denoising and enhancing the RGB image which is originally input by utilizing a bilateral filter; Converting each pixel value in the image after bilateral filtering into a CIELAB color space with uniform perception to obtain an original brightness component and an original chromaticity component; Performing nonlinear tone mapping on the luminance component using Gamma correction techniques; Processing the image after Gamma correction processing by adopting a limited contrast self-adaptive histogram equalization method to obtain a local enhanced brightness component; introducing a Gaussian differential technology to directionally enhance the intermediate frequency component of the image; and pulling the overall brightness mean value of the image subjected to directional enhancement to a statistical distribution center through dynamic brightness anchoring, outputting a final brightness channel, and finally reconstructing the final brightness component and the chromaticity component back to an RGB space.
2. The illumination adaptive image preprocessing method for Shan Mugao S sputter SLAM as recited in claim 1, characterized in that for any pixel point in the original input RGB image Its filtered output value The definition is as follows: ; Wherein, the To take the following measures Is a neighborhood window that is centered, Is the pixel point in the neighborhood Is used for the original gray-scale value of (c), Is a pixel point Is used for the original gray-scale value of (c), Representing the Euclidean distance; for normalizing coefficients, ensure that the sum of filter kernel weights is 1: ; The formula contains two Gaussian kernel functions and a space kernel For measuring Euclidean distance between pixels, value domain kernel For measuring the gray scale difference between pixels.
3. The illumination adaptive image preprocessing method for Shan Mugao s sputtering SLAM of claim 1, wherein the specific processing procedure of Gamma correction is as follows: output brightness Equal to a constant Multiplying by input brightness A kind of electronic device Power of: ; Wherein (x, y) is the coordinate value of the pixel point, a series of brightness demarcation point sets are preset And corresponding object thereof Value set Constructing luminance channel mean values for pixels of an input image using piecewise linear functions And correction coefficient Continuous mapping relation between : ; Wherein i represents the index of the brightness demarcation points, n represents the number of the brightness demarcation points, Is a linear interpolation function.
4. The illumination adaptive image preprocessing method for Shan Mugao s sputtered SLAM of claim 1, wherein the limiting contrast adaptive histogram equalization method is implemented as follows: Dividing the image after Gamma correction into segments The sub-blocks are not overlapped with each other, and a histogram is calculated for each sub-block independently; for each sub-block, its gray level histogram is first calculated Introducing a clipping threshold to limit excessive enhancement When a certain gray level Frequency of (2) Exceeding the limit When the pixel is cut off, the sum of all cut-off pixels is counted : ; Wherein the method comprises the steps of Is the total number of gray levels, overflowed pixels obtained by statistics Evenly distributed over all gray levels of the histogram, resulting in a corrected histogram : ; Based on the corrected histogram, calculating a cumulative distribution function as a gray mapping curve, the corrected gray transformation being described as: ; Wherein the method comprises the steps of Is based on The cumulative probability distribution calculated after normalization, And Finally, adopting bilinear interpolation technology to carry out weighted fusion on any pixel in the image according to the position of the pixel by utilizing the transformation functions of the adjacent four sub-blocks so as to ensure the smooth transition of gray level transformation in the global range; finally, after full-image interpolation and fusion, outputting local enhanced brightness components : ; Wherein the method comprises the steps of Representing the complete mapping operator including statistics, clipping, reassignment and interpolation, For the sub-block grid size, The brightness of the pixel point with coordinates (x, y), i.e. the corrected gray scale, Represents clipping thresholds constrained by a combination of a luminance constraint factor, a contrast compensation factor, and a texture suppression factor.
5. The illumination-adaptive image preprocessing method for Shan Mugao s sputtered SLAM as set forth in claim 4, characterized in that the luminance constraint factor According to local brightness mean Dynamically adjusting the mapping relation expressed as linear interpolation form by using function ) The representation is: ; Wherein the method comprises the steps of Respectively representing the sets before and after mapping, wherein n is the number of elements in the sets; Contrast compensation factor Using standard deviation of brightness of all pixels of an image To measure the overall contrast of the image: 。
6. The illumination-adaptive image preprocessing method for Shan Mugao S sputter SLAM as recited in claim 4, characterized by a texture suppression factor Dynamically adjusting the enhancement upper limit according to the texture richness of the image, in particular, utilizing a second order differential operator Laplacian To extract the high frequency edge information of the luminance channel: ; Wherein, the Representing the brightness of the pixel with coordinates (x, y), and then calculating the global variance : ; Wherein, the To be the mean of the laplace response plot, And The height and width of the image are respectively; Finally, the specific self-adaptive mapping relation adopts piecewise linear interpolation function The realization is as follows: ; Wherein, the Respectively representing the sets before and after mapping, and n is the number of elements in the set.
7. The illumination-adaptive image preprocessing method for Shan Mugao S sputter SLAM as recited in claim 4, characterized by a final clipping threshold Is a luminance constraint factor Contrast compensation factor And texture suppression factor And undergo a numerical truncation process to ensure numerical stability: ; Wherein, the And Are all constant.
8. The illumination adaptive image preprocessing method for Shan Mugao s sputtered SLAM as set forth in claim 1, wherein the Gaussian difference technique is defined by the difference between convolutions of two different scale Gaussian kernels: ; wherein the gaussian kernel function Is a standard gaussian distribution, the distribution is a standard gaussian distribution, And Respectively the size of two Gaussian kernel functions, and the enhanced image The method is formed by fusing a basic brightness layer and a weighted Gaussian difference layer: ; Fusion weights Standard deviation from image brightness Negative correlation is formed; dynamic luminance anchoring is based on the current luminance mean Calculating deviation from target mean : ; Finally output brightness channel The method comprises the following steps: ; Wherein, the Is constant.
9. An illumination adaptive image preprocessing system for Shan Mugao s sputter SLAM, characterized by comprising a processor and a memory for storing program instructions, the processor for invoking the program instructions in the memory to perform the illumination adaptive image preprocessing method for Shan Mugao s sputter SLAM as set forth in any one of claims 1-8.
10. A computer readable storage medium comprising a readable storage medium having stored thereon a computer program which, when executed, implements the illumination adaptive image preprocessing method for Shan Mugao s sputter SLAM according to any one of claims 1-8.

Description

Illumination self-adaptive image preprocessing method and system for Shan Mugao-step sputtering SLAM Technical Field The invention belongs to the technical field of computer vision and robots, and particularly relates to an illumination self-adaptive image preprocessing method and system for single-eye Gaussian sputtering SLAM. Background At the moment of rapid development of automation and intelligent technology, a visual synchronous positioning and mapping technology (SLAM) becomes a key technology for endowing a machine with autonomous sensing and exploration of an unknown environment, and is widely applied to the front edge fields of automatic driving, intelligent self-body, digital twin systems and the like. In recent years, three-dimensional Gaussian sputtering technology (3D Gaussian Splatting, 3 DGS) has shown higher view synthesis quality by virtue of anisotropic Gaussian ellipsoid explicit expression scene and matching with differentiable real-time rendering pipelines, and provides a brand new technical paradigm for constructing a dense map with centimeter-level positioning accuracy and photo-level fidelity. However, whether classical sparse feature point-based geometric methods such as ORB-SLAM, VINS-Mono, or emerging 3DGS systems based on dense photometric errors, their robust operation tends to build on two idealized assumptions of photometric consistency and static scene. In the face of severe indoor and outdoor illumination changes, dynamic shadow interference, extremely low illuminance environments and the like, nonlinear illumination disturbance can cause automatic exposure adjustment, so that the description of image characteristic points is severely drifted, deviation is directly caused in characteristic extraction and matching links, further, the problems of tracking loss and rapid increase of accumulated errors of a front-end visual odometer are caused, and artifacts, geometric cavities and the like are caused in a reconstructed map. Strategies to cope with illumination changes are mainly embodied at two levels, namely feature representation and system robustness enhancement. At the feature level, researchers use local descriptors (such as ORB, LIFT, etc.) or fast binary features with illumination invariance to promote the stability of feature matching under illumination changes. On the system level, on one hand, the robustness of closed loop detection is enhanced by introducing semantic information (such as semantic word bags and scene level descriptors) and combining geometric constraints, and on the other hand, a semantic topological structure of a static object is constructed by means of target detection and the like, and relocation is assisted by topological similarity, so that the stability of the system under long-term operation and severe illumination change is improved. However, these methods are essentially passive compensation mechanisms in that once the texture information has been lost in the imaging stage due to severe overexposure or underexposure of the input image, subsequent optimizations tend to be difficult to recover the valid information. Even advanced systems that introduce geometric uncertainty are not modeled specifically for nonlinear illumination changes, illumination abrupt changes can still significantly interfere with the accuracy of uncertainty estimation. Therefore, it is necessary to construct a monocular SLAM dual-stage illumination self-adaptive preprocessing frame, and illumination robustness is improved on the premise of keeping the real-time performance of the system. Disclosure of Invention The invention aims to solve the problems that the front end seriously depends on a luminosity consistency assumption when the existing 3D Gaussian sputtering SLAM system faces severe illumination fluctuation, and positioning drift and image creation artifacts are easy to generate, and provides an illumination self-adaptive preprocessing method capable of enhancing system robustness in an illumination abrupt change and extremely dark environment. In order to achieve the above purpose, the present invention provides an illumination adaptive pretreatment method for single-mesh gaussian sputtering SLAM, which comprises the following specific steps: Denoising and enhancing the RGB image which is originally input by utilizing a bilateral filter; Converting each pixel value in the image after bilateral filtering into a CIELAB color space with uniform perception to obtain an original brightness component and an original chromaticity component; Performing nonlinear tone mapping on the luminance component using Gamma correction techniques; Processing the image after Gamma correction processing by adopting a limited contrast self-adaptive histogram equalization method to obtain a local enhanced brightness component; introducing a Gaussian differential technology to directionally enhance the intermediate frequency component of the image; and pulling the overall brightness m