WO2018023916A1 - 一种彩色图像去阴影方法和应用 - Google Patents

一种彩色图像去阴影方法和应用 Download PDF

Info

Publication number
WO2018023916A1
WO2018023916A1 PCT/CN2016/109612 CN2016109612W WO2018023916A1 WO 2018023916 A1 WO2018023916 A1 WO 2018023916A1 CN 2016109612 W CN2016109612 W CN 2016109612W WO 2018023916 A1 WO2018023916 A1 WO 2018023916A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
feature
unshaded
shadow
road surface
Prior art date
Application number
PCT/CN2016/109612
Other languages
English (en)
French (fr)
Inventor
李革
应振强
Original Assignee
北京大学深圳研究生院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京大学深圳研究生院 filed Critical 北京大学深圳研究生院
Priority to US16/089,462 priority Critical patent/US10592754B2/en
Publication of WO2018023916A1 publication Critical patent/WO2018023916A1/zh

Links

Images

Classifications

    • G06T5/94—
    • G06T5/70—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/60—Extraction of image or video features relating to illumination properties, e.g. using a reflectance or lighting model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle
    • G06T2207/30256—Lane; Road marking

Definitions

  • the invention relates to the field of machine vision, in particular to a color image shading method and an application for detecting a road pavement area based on the method.
  • the shadow in the image has an effect on the edge extraction and the machine vision algorithm of the image matching.
  • the color image is composed of three RGB gray component images. Since the grayscale images of these three components are very similar, machine vision algorithms usually convert them to a grayscale component for processing. Since the luminance information is included in all of the three gray components, the interference of the illumination changes is directly caused by the three components or their weighted sum and maximum and minimum components. Especially in the case of shadows, the effects of illumination on the machine vision algorithm are very large.
  • the existing methods for shadow preprocessing are mainly divided into anti-shadow feature extractor, unshadow feature extractor and shadow detection and removal methods:
  • Anti-shadow feature extractor Color components are extracted by color space conversion and color components that are insensitive to illumination changes, and color components are used for subsequent processing. Since the description of the color space is an ideal model, although these extracts can reduce the interference of the shadow to a certain extent, the shadows in the image cannot be completely removed. Especially in the case of strong shadows, these extractors may fail completely.
  • Shadowless feature extracts obtained by imaging physics theory analysis and experiments can completely remove shadows.
  • the existing methods still cannot solve some strong shadow situations.
  • Shadow detection and removal methods The shadow area is first detected, and then the image of the shadow portion is repaired using a recovery algorithm. This method tends to be more complex and slower to process, and most existing methods lose the details of the shadow area to varying degrees.
  • the existing shadow processing method can not achieve good shadow removal effect while ensuring real-time performance (fast processing speed), especially in the case of strong shadow.
  • Intelligent vehicle sensing technology enables vehicles to actively sense the surrounding environment, thereby actively preventing traffic accidents and even completing automatic driving.
  • Road recognition is a prerequisite for driving, so road detection is an integral part of smart vehicle perception.
  • the road detection method processes a road image of a driver's perspective to detect the area in which the road surface is located. Trees on the roadside, buildings will leave shadows on the road, especially in the case of strong light, the interference of the shadows is very serious, making it difficult to detect the road area correctly.
  • the present invention provides a color image shading method and an application for detecting a road pavement area based on the method, and the method of the present invention can remove the shadow existing in the color image, and is applied as a pre-processing step.
  • a variety of machine vision areas The application for detecting the road pavement area based on the method specifically utilizes the de-shadowing method to solve the road pavement detection problem under strong shadow.
  • the image de-shadowing method provided by the invention comprises unshaded feature analysis, non-shadow transform parameter acquisition, and unshaded feature imaging;
  • the road pavement detection method comprises first performing region selection and feature extraction (using the previous unshaded feature extraction method), Image filtering, segmentation, and road surface selection are performed, followed by image morphology filtering and hole filling.
  • the color image deshaping method provided by the invention comprises a shadowless feature analysis process, a shadowless transform parameter acquisition process and a shadowless feature imaging process.
  • the image deshaping method described above can be applied to road detection; in addition to road detection, it can be applied to other scenes, including foreground segmentation, image matching, color-based image segmentation, and edge extraction.
  • the foreground segmentation algorithm is applied to reduce the interference caused by the shadow.
  • Image Matching the difference in illumination is also a major problem in image matching. We hope that objects in different illuminations can be correctly matched, and it is not desirable to invalidate the matching algorithm due to uneven illumination caused by local shadows. Therefore, the method is also applicable to image matching problems, and further, can be applied to image-based positioning systems.
  • An image de-shadowing method includes an unshaded feature analysis extraction process, a non-shadow transformation parameter acquisition process, and a non-shadow feature imaging process; and specifically includes the following steps:
  • Unshadowed feature imaging process converting the unshaded matrix into a grayscale image, and obtaining a shadowless feature grayscale image by feature visualization.
  • step 1) defines an expression of the unshaded feature by using a linear relationship of RGB spaces; the linear relationship of the RGB space includes a linear relationship of RGB every two components or RGB three-points The linear relationship of the linear combination of quantities.
  • step 1) selects a linear relationship of the B and G components to represent a non-shadow feature of the image for road surface detection, and the expression of the unshaded feature is Equation 1:
  • the geometric meaning is the intercept of the fitted straight line; for the pavement area, the value of K is a fixed value, and K is the unshaded feature matrix sought.
  • the step 3) converts the unshaded feature matrix into a grayscale image, and specifically extracts the grayscale value of the material of the region of interest by the image segmentation method.
  • the image segmentation method specifically normalizes the unshadowed feature matrix such that the values of the elements of the matrix are distributed between 0 and 1, thereby obtaining a shadowless feature grayscale image. More specifically, the lower bound and the upper bounds m and n of the optimal interval of the natural image feature value are obtained by solving the optimal partition interval method, and then the interval (m, n) is mapped to the interval (0, 1) for normalization. Obtaining a shadowless grayscale image C; the optimal partitioning interval is solved by Equations 1 to 3:
  • K corresponding to the pixel values in one or more images are respectively calculated, and the maximum value and the minimum value of K values are respectively recorded as K max and K min , and the statistical histogram is recorded as H:
  • the final unshaded grayscale image C is obtained.
  • the image deshaping method provided by the present invention can be applied to various machine vision fields, including image foreground segmentation, image matching, color-based image segmentation and edge extraction, and road detection.
  • the present invention specifically provides an application method of the above image de-shadowing method in road detection, including an image pre-processing process, a road surface extraction process, and an image post-processing process; specifically, the region selection and the region of interest are first selected by a non-shadow feature analysis extraction process.
  • Feature extraction, image filtering, segmentation and road surface selection, and finally image morphological filtering and holes Hole filling specifically includes the following steps:
  • a road surface extraction process including image segmentation and road surface region selection; dividing the road image into a plurality of regions by the image segmentation; and evaluating each region by the road surface region selection, and selecting an area most likely to be a road surface;
  • Image post-processing removing the excess part by image morphology filtering; filling the holes in the pavement area with a hole filling algorithm;
  • step 81) specifically removes the irrelevant region by selecting the region of interest; and removes the interference of the shadow by the unshaded feature extraction method to obtain an unshaded grayscale image.
  • C is a shadowless grayscale image
  • G and B respectively correspond to the component matrix of the green and blue colors of the image
  • b is a camera parameter, and the intercept of the straight line on the G axis is obtained for fitting.
  • the method Compared with the three types of prior art summarized in the background art (anti-shadow feature extractor, unshaded feature extractor and shadow detection and removal method), the method combines the advantages of low complexity (determining processing speed) and high accuracy. . For some cases of strong shadows, both the anti-shadow feature extractor and the unshaded feature extractor do not work well. Moreover, the feature images extracted by the anti-shadow feature extractor and the unshadow feature extractor have poor discrimination, which is not conducive to subsequent image processing and analysis work.
  • the de-shadowing method provided by the present invention has a good degree of discrimination and facilitates subsequent processing such as image segmentation.
  • the third type of technical shadow detection and removal defects are high in complexity and slow in processing speed, and the method of the present invention achieves an accurate deshaping effect in the case of low complexity, and can cope with situations where real-time requirements are high.
  • the method of the present invention is capable of removing shadows present in color images as a pre-processing step for a variety of machine vision fields. Firstly, the pre-processing method provided by the present invention is used to perform pre-processing, and then the corresponding processing method can be used to improve the robustness against shadow interference and achieve better performance.
  • FIG. 1 is a schematic diagram showing a distribution rule of pixel values of the same material plane in an RGB space in an embodiment of the present invention
  • FIG. 2 is a view showing a shading effect according to an embodiment of the present invention.
  • the left picture is the gray image of the original image; the right picture is the corresponding unshaded image.
  • Figure 3 is a road detection task description
  • the left picture (a) is the original image as an input; the right picture (b) is the output processing result, and the position of the road surface area is represented by a binary image.
  • FIG. 4 is a flow chart of a method for applying an image deshaping method to road surface detection provided by the present invention.
  • the invention provides a color image shading method and an application for detecting a road pavement area based on the method, and provides an implementation example of solving the road pavement detection problem under strong shadow by using the deshaping method.
  • the method of the present invention is capable of removing shadows present in color images and applying them as a pre-processing step to a variety of machine vision fields.
  • the method of de-shadowing includes unshaded feature analysis, unshadowed parameter acquisition, and unshaded feature imaging.
  • the road pavement detection method includes first performing region selection and feature extraction (using the previous unshaded feature extraction method), and then performing image filtering. Segmentation and pavement area selection, and finally image morphology filtering and hole filling.
  • the design framework of the unshaded feature extractor consists of three major steps: unshaded feature analysis, unshadowed transform parameter acquisition, and unshaded feature imaging:
  • Unshaded feature analysis According to the statistical analysis of the RGB values of the same material under different illumination in the shadowed image or the physical imaging model, the invariant characteristic law is found, and the description of the law is the non-shadow feature.
  • K f(R, G, B,%), where K is the unshaded feature matrix, R, G, B corresponds to the red, green and blue color component matrix of the original image, and the ellipsis indicates other possible parameters, f
  • K the unshaded feature matrix
  • R, G, B corresponds to the red, green and blue color component matrix of the original image
  • the ellipsis indicates other possible parameters
  • f To convert the expression.
  • Figure (c) For different problems, we can use different definitions. Taking the road surface detection problem as an example, we can select only the B and G components and give the following definitions without
  • No shadow transformation parameter acquisition in addition to the pixel values (R, G, B) of each point to be determined, there are generally other parameters that need to be determined. These parameters are determined by data fitting or other calibration methods.
  • b is the parameter to be determined, and its geometric meaning is the intercept of the fitted line.
  • the plane we are interested in for the road surface detection problem is the road surface.
  • the pixel of the road surface area is fitted, and the value of the intercept of the obtained straight line on the G axis is taken out, which is the value of b to be determined.
  • After determining the b value enter a road color image (RGB known) and we can calculate the corresponding unshaded matrix.
  • Shadowless feature imaging This step is to visualize the obtained features (convert the matrix into a grayscale image suitable for human eye observation), and the feature is visualized to obtain a shadowless feature grayscale image.
  • the gray value of the material of the region of interest is consistent and can be extracted by the image segmentation method.
  • the simplest method of imaging is to normalize the feature matrix directly, that is, to distribute the values of the elements of the matrix between 0 and 1. If you have a very large or very small pixel, the overall image contrast will be low. Therefore, in order to obtain an effective feature grayscale image, we need to analyze the distribution of the eigenvalues of the natural image for the application, and eliminate those small or small pixel values.
  • the pixel values in one or more road images can be taken out, and the values of K corresponding to the pixel values are respectively calculated, and the maximum and minimum values of K values are respectively recorded as K max and K min .
  • the statistical histogram is recorded as H, and the solution of the optimal partition interval can be defined as an optimization problem:
  • the invention also provides a road detection framework based on unshaded feature extraction, and FIG. 1 is an image provided by the present invention.
  • the flow chart of the method of applying the shadow method to road pavement detection mainly includes three parts:
  • Image preprocessing The purpose of the pre-processing is to take out irrelevant regions and interference information, where the irrelevant regions are removed by the region of interest selection, and the interference of the shadows is removed by the unshaded feature extraction;
  • Road extraction After pre-treatment, the existing road surface extraction method can resist the interference of shadows and no longer fails under strong shadows.
  • Road surface extraction is mainly divided into two major steps: image segmentation and road surface selection. Dividing the road image into regions by image segmentation; then selecting the pavement region: evaluating each region and selecting the region most likely to be the road surface;
  • Pavement candidate areas obtained from road surface extraction are not perfect and may contain excess or missing areas.
  • image morphological filtering the excess portion can be removed; the lane markings on the road may form holes in the area, and these holes are filled by a hole filling algorithm. After the post-processing is completed, the final road test result is output.
  • Embodiment 1 is a diagrammatic representation of Embodiment 1:
  • a shadowless feature extractor suitable for road surface detection is provided, and the steps are as follows:
  • Offline camera calibration The image of the application scene is calibrated, that is, the object area in which the object is of interest is manually marked. Calibration work can be done using interactive software. For the pavement detection, the area we are interested in is the pavement. Open a shadowed picture and circle some of the pavement area. As shown in Figure 1(a), the interactive software will image the pixels inside the area, as shown in Figure 1 (b). ), drawn to the RGB space as shown in Figure 1 (c), and then projected to the GB space, the distribution of the pixel point cloud can be fitted by a straight line equation, the software fits the data to get the intercept of the line on the G axis b is the required camera parameters, as shown in Figure 1 (d).
  • the mapping range is determined.
  • the color road image can be converted into a shadowless grayscale image, and the experimental results show that the formula can be applied not only to road images, but also to most natural images, the conversion formula can be well de-shaded. effect.
  • the effect of the unshaded feature extraction obtained by this method is shown in the figure.
  • Embodiment 2 is a diagrammatic representation of Embodiment 1:
  • a road pavement detecting method processes a road image of a driver's perspective to detect the area in which the road surface is located, as shown in FIG.
  • the method framework of the road pavement detection is shown in FIG. 4 As shown, it specifically includes the following steps:
  • Region of interest Region of interest selection. This step limits subsequent processing steps to a sub-area to increase system efficiency. Processing unrelated areas not only wastes time, but may also introduce noise that affects the performance of the system. So the first step is to define which image areas are useless and which areas are useful for implementing the application. The area useful for subsequent processing is called the area of interest.
  • the area useful for subsequent processing is called the area of interest.
  • From a simple point of view generally select the region of interest of the rectangle, according to the rectangular ROI window, directly obtain a sub-image, which facilitates subsequent processing.
  • the road detection problem we are interested in the pavement part. In the road image of the collection equipment (such as driving recorder, car camera), the road surface is generally in the lower part, while the upper part is mainly the sky, which is interference to the road detection. information. Simply, the lower half of the image can be directly selected as the region of interest, or it can be adjusted according to the actual application, and a more accurate ROI is adopted.
  • Pavement feature extraction The color information is generally used for the division of the road surface, but the division by the color information may be invalid due to the possibility of a large amount of shadow on the road surface.
  • Image filtering denoising There may be noise in the original image, and the previous road feature extraction also introduces noise. This step removes these noises by applying a set of filters to the image. First, median filtering is used to remove the salt and pepper noise, and secondly, smoothing filtering is used to filter out high frequency noise.
  • Image segmentation The image segmentation algorithm is used to segment the filtered image into two connected regions.
  • Pavement area selection Since the road surface is the most prominent area in the ROI area, the occupied area is also the largest, and the road area with the largest area can be extracted by the analysis of the connected components.
  • the use of connected components is the easiest way to screen out the pavement area in the segmentation area. It is also possible to select the area closest to the road feature by classifying the regional features, thereby extracting the road surface more accurately.
  • the pavement area obtained by the connected component analysis may contain some extraneous regions, and the morphological filtering of the binary images can remove these extraneous regions.
  • the road markings on the road surface may form a number of holes in the extracted area, and the holes are filled by the hole filling algorithm to obtain the final road surface inspection result.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

公布了一种图像去阴影方法和应用。去阴影方法包括无阴影特征分析过程、无阴影变换参数获取过程和无阴影特征成像过程。应用于道路路面检测，检测方法包括先采用无阴影特征提取方法进行感兴趣区域选择和特征提取，再进行图像滤波、分割和路面区域选择，最后进行图像形态学滤波和孔洞填充。该方法能够去除彩色图像中存在的阴影，作为预处理步骤应用到多种机器视觉领域，道路路面检测应用可解决强阴影下的道路路面检测问题，具有低复杂度、处理速度快和高准确度的优势。

Description

一种彩色图像去阴影方法和应用 技术领域
本发明涉及机器视觉领域，尤其涉及一种彩色图像去阴影方法和基于该方法对道路路面区域进行检测的应用。
背景技术
图像中的阴影对边缘提取，图像匹配的机器视觉算法都有影响，彩色图像由RGB三个灰度分量图像构成。由于这三个分量的灰度图像非常类似，机器视觉算法通常先将它们转为一个灰度分量进行处理。由于这三个灰度分量中都包含亮度信息，所以直接采用这三个分量或者他们的加权和、最大最小值构成的灰度分量都存在光照变化的干扰问题。特别是在阴影情况下，光照变化对机器视觉算法的干扰很大。现有方法对阴影的预处理主要分为抗阴影特征提取子、无阴影特征提取子以及阴影检测和去除方法三种：
(一)抗阴影特征提取子。通过颜色空间转换将亮度成分和对光照变化不敏感的颜色成分抽取出来，采用颜色分量进行后续处理。由于颜色空间的描述是理想模型，这些提取子虽然可以一定程度地减弱阴影的干扰，但无法完全地去除图像中的阴影。特别是在强阴影情况下，这些提取子可能完全失效。
(二)无阴影特征提取子。通过成像物理理论分析和实验得出的无阴影特征提取子可以完全地去除阴影。但现有的方法仍然无法解决一些强阴影情形。
(三)阴影检测和去除方法。首先对阴影区域进行检测，然后再利用恢复算法对阴影部分的图像进行修复。这种方法往往复杂度较高，处理速度较慢，且绝大多数的现有方法都会不同程度地丢失阴影区域的细节信息。
综合来看，现有对阴影的处理方法无法在保证实时性的同时(处理速度快)达到很好的阴影去除效果，特别是在强阴影情形下。这就使得阴影干扰成为很多实时性要求高的机器视觉任务的难点，比如智能车辆感知技术中的道路检测问题。
智能车辆感知技术使得车辆能够主动地感知周围环境，进而积极地预防交通事故，甚至完成自动驾驶。认路是驾驶的前提，因此道路检测是智能车辆感知不可或缺的一部分。道路检测方法对一张驾驶员视角的道路图像进行处理，检测出其中路面所在的区域。路边的树木，建筑物都会在道路上留下阴影，特别是在光照强烈的情形，阴影的干扰十分严重，使得道路区域难以正确地检出。
发明内容
为了克服上述现有技术的不足，本发明提供一种彩色图像去阴影方法和基于该方法对道路路面区域进行检测的应用，本发明方法能够去除彩色图像中存在的阴影，作为预处理步骤应用到多种机器视觉领域。基于该方法对道路路面区域进行检测的应用具体利用所述去阴影方法解决强阴影下的道路路面检测问题。
本发明提供的图像去阴影方法包括无阴影特征分析、无阴影变换参数获取、无阴影特征成像；道路路面检测方法包括先进行感兴趣区域选择和特征提取(采用前面的无阴影特征提取方法)，再进行图像滤波、分割和路面区域选择，最后进行图像形态学滤波和孔洞填充。本发明提供的彩色图像去阴影方法，包括无阴影特征分析过程、无阴影变换参数获取过程和无阴影特征成像过程。上述图像去阴影方法可应用于道路检测，；除道路检测外，还可应用于其他场景，包括前景分割、图像匹配、基于颜色的图像分割和边缘提取等。在前景分割问题中，由于感兴趣的前景物体的投影也会随着物体的运动而运动，同时背景中也有可能有运动的阴影，这会使得部分阴影被错误地识别为前景。应用该方法去除阴影后，再应用前景分割算法，能够减弱阴影造成的干扰。图像匹配此外，光照差异也是图像匹配的一大问题，我们希望在不同光照下的物体能够正确的匹配，而不希望由于局部阴影造成的光照不均匀使得匹配算法失效。因此本方法也适用于图像匹配问题，进一步地，可以应用到基于图像的定位系统中。
本发明提供的技术方案是：
一种图像去阴影方法，包括无阴影特征分析提取过程、无阴影变换参数获取过程和无阴影特征成像过程；具体包括如下步骤：
1)无阴影特征分析提取过程：对原始图像中同一材料在不同光照下的像素RGB值进行统计分析，或者通过物理成像模型找出含阴影图像中不变的特性，获得无阴影特征的表达式K＝f(R,G,B,…)，其中，K为无阴影特征矩阵；R、G、B分别对应原始图像的红绿蓝颜色的三个分量矩阵；…表示其他额外参数；f为转换表达式；
2)无阴影变换参数获取过程：对步骤1)所述无阴影特征矩阵K＝f(R,G,B,…)，通过确定表达式中的参数，计算得到对应的无阴影特征矩阵K；
3)无阴影特征成像过程：将所述无阴影矩阵转换为灰度图像，通过特征可视化得到无阴影特征灰度图像。
针对上述图像去阴影方法，进一步地，步骤1)利用RGB空间的线性关系定义所述无阴影特征的表达式；所述RGB空间的线性关系包括RGB每两个分量的线性关系或RGB三分 量的线性组合的线性关系。
针对上述图像去阴影方法，进一步地，步骤1)选用B和G分量的线性关系表示用于路面检测的图像的无阴影特征，所述无阴影特征的表达式为式1：
Figure PCTCN2016109612-appb-000001
其中，b为待确定的其他参数，几何意义是拟合直线的截距；对于路面区域，K的取值为一个定值，K即为所求的无阴影特征矩阵。
针对上述图像去阴影方法，进一步地，步骤3)所述将无阴影特征矩阵转换为灰度图像，具体通过图像分割方法提取得到感兴趣区域材料的灰度值。更进一步地，所述图像分割方法具体采用对无阴影特征矩阵进行归一化，使得矩阵各个元素的值分布在0和1之间，从而得到无阴影特征灰度图像。更具体地，通过求解最优划分区间的方法获得自然图像特征值最优区间的下界和上界m和n，再将区间(m,n)映射到区间(0，1)做归一化，得到无阴影灰度图C；所述最优划分区间通过式1～式3求解：
分别计算得到一张或多张图像中的像素值对应的K的取值，将K取值的最大值和最小值分别记作Kmax和Kmin，统计直方图记作H：
max:g(m,n)-c(m,n)          (式1)
Figure PCTCN2016109612-appb-000002
Figure PCTCN2016109612-appb-000003
其中，m和n为所求最优区间的下界和上界；将(m,n)区间映射到(0，1)区间做归一化：
Figure PCTCN2016109612-appb-000004
得到最终的无阴影灰度图C。
本发明提供的上述图像去阴影方法可应用于多种机器视觉领域，包括图像前景分割、图像匹配、基于颜色的图像分割和边缘提取、道路检测等场景。
本发明具体提供了上述图像去阴影方法在道路检测中的应用方法，包括图像预处理过程、路面提取过程和图像后处理过程；具体地，首先通过无阴影特征分析提取过程进行感兴趣区域选择和特征提取，再进行图像滤波、分割和路面区域选择，最后进行图像形态学滤波和孔 洞填充；具体包括如下步骤：
81)进行图像预处理，取出无关区域和干扰信息；
82)路面提取过程：包括图像分割和路面区域选择；通过所述图像分割，将道路图像划分成多个区域；通过所述路面区域选择对各个区域进行评估，选择得到最可能为路面的区域；
83)图像后处理过程：通过图像形态学滤波，将多余的部分去除；采用孔洞填充算法对路面区域中的孔洞进行填补；
完成后处理后，输出道路检测结果。
针对上述图像去阴影方法在道路检测中的应用，其中，步骤81)具体通过选择感兴趣区域去除无关区域；通过无阴影特征提取方法去除掉阴影的干扰，得到无阴影灰度图。
进一步地，所述无阴影灰度图为式5：
Figure PCTCN2016109612-appb-000005
式5中，C为无阴影灰度图；G、B分别对应图像的绿和蓝颜色的分量矩阵；b为相机参数，为拟合得到直线在G轴上的截距。
与现有技术相比，本发明的有益效果是：
相比背景技术中总结的三类现有技术(抗阴影特征提取子、无阴影特征提取子和阴影检测和去除方法)，本方法结合了低复杂度(决定处理速度)和高准确度的优势。对于一些强阴影的情形，抗阴影特征提取子和无阴影特征提取子都无法很好地工作。而且，抗阴影特征提取子和无阴影特征提取子提取出的特征图像区分度差，不利于后续的图像处理和分析工作。
相比之下，本发明提供的去阴影方法具有很好的区分度，便于图像分割等后续处理。第三类技术阴影检测和去除的缺陷是复杂度高，处理速度慢，而本发明所述的方法在低复杂度的情况下取得了准确的去阴影效果，能够应对实时性要求高的场合。因此，本发明方法能够去除彩色图像中存在的阴影，作为预处理步骤应用到多种机器视觉领域。先采用本发明提供的去阴影方法进行预处理，之后再使用相应处理方法，可以提高其对阴影干扰的鲁棒性，取得更好的性能。
附图说明
图1是本发明实施例中采用同一材料平面的像素值在RGB空间的分布规律示意图；
其中，(a)为从道路图像中选出的路面区域(感兴趣对象的样本)；(b)为该区域的各个像素点；(c)为将像素点绘制到RGB三维空间，这些点分布在一条直线附近；(d)为将三维空间的像素点投影到GB二维平面上，这些点的坐标可以用一个直线去拟合。
图2为本发明实施例的去阴影效果图；
其中，左图均为原始图像的灰度图；右图为得到的相应的无阴影图像。
图3是道路检测任务说明；
其中，左图(a)是原始图像，作为输入；右图(b)为输出的处理结果，通过一个二值图像表征路面区域位置。
图4是本发明提供的将图像去阴影方法应用于道路路面检测的方法流程框图。
具体实施方式
下面结合附图，通过实施例进一步描述本发明，但不以任何方式限制本发明的范围。
本发明提供一种彩色图像去阴影方法和基于该方法对道路路面区域进行检测的应用，提供了利用所述去阴影方法解决强阴影下的道路路面检测问题的实施实例。本发明方法能够去除彩色图像中存在的阴影，作为预处理步骤应用到多种机器视觉领域。
以下叙述本发明提供的图像去阴影方法和道路路面检测方法的具体实施。去阴影方法包括无阴影特征分析，无阴影变换参数获取、无阴影特征成像；道路路面检测方法包括先进行感兴趣区域选择和特征提取(采用前面的无阴影特征提取方法)，再进行图像滤波、分割和路面区域选择，最后进行图像形态学滤波和孔洞填充。
在图像去阴影方法中，无阴影特征提取子的设计框架包括三大步骤：无阴影特征分析，无阴影变换参数获取和无阴影特征成像：
无阴影特征分析：根据对含阴影图像中同一材料在不同光照下的像素RGB值进行统计分析或者从物理成像模型出发，找出其中不变的特性规律，对该规律的描述即为无阴影特征的表达式K＝f(R,G,B,…)，其中K为无阴影特征矩阵，R，G，B对应原始图像的红绿蓝三个颜色分量矩阵，省略号表示其他可能的参数，f为转换表达式。具体地，我们采用RGB空间的线性规律：把位于同一材料平面上的像素映射到RGB空间，它们的像素值分布在一条直线上，如图(c)。对于不同的问题，我们可以采用不同的定义方式，以路面检测问题为例，我们可以仅选用B,G分量，给出下列无阴影特征定义：
Figure PCTCN2016109612-appb-000006
其中，b为待确定的额外参数。该定义的导出是利用了B,G分量之间的线性关系。将图右图的三维像素分布投影到BG坐标平面后，可以看到BG之间服从线性关系(如图1(d))，因此对于路面区域，K的取值为一个定值，K即为所求的无阴影特征矩阵。此外，利用RGB空间的线性规律还可以定义很多其他的变式，如选用其他两个分量的线性关系，或利用三分量的线性组合之间的线性关系。
无阴影变换参数获取。在表达式中，除了各点的像素值(R、G、B外)待确定外，一般还有其他参数需要确定。通过数据拟合或者其他标定方法确定这些参数。在前面的定义中，b是待确定的参数，它的几何意义是拟合直线的截距。对于路面检测问题我们感兴趣的平面是路面，对路面区域像素做拟合，取出所得直线在G轴上的截距的值，即为待确定的b的值。确定了b值后，输入一张道路彩色图像(RGB已知)，我们就可以计算出对应的无阴影矩阵。
无阴影特征成像。这一步进行对所得特征的可视化(将矩阵转换为适合人眼观察的灰度图像)，特征可视化后就得到一个无阴影特征灰度图像。在得到的无阴影图像中，感兴趣区域材料的灰度值一致，可以通过图像分割方法提取出来。最简单的成像方法是对特征矩阵直接进行归一化，即使得矩阵各个元素的值分布在0和1之间。如果有一个取值非常大或者非常小的像素，就会使得整个图像对比度很低。因此，为了得到有效的特征灰度图像，我们需要分析对于应用场合的自然图像该特征值的分布情况，剔除那些少数偏大或者偏小的像素值。以路面检测问题为例，可以取出一张或多张道路图像中的像素值，分别计算出这些像素值对应的K的取值，将K取值的最大最小值分别记作Kmax和Kmin，统计直方图记作H，则最优划分区间的求解可以定义为一个最优化问题：
max:g(m,n)-c(m,n),
Figure PCTCN2016109612-appb-000007
Figure PCTCN2016109612-appb-000008
其中，m和n为所求最优区间的下界和上界。求得最优解后，将(m,n)区间映射到(0，1)区间做归一化，就得到最终的无阴影灰度图C：
Figure PCTCN2016109612-appb-000009
本发明还提供一种基于无阴影特征提取的道路检测框架，图1是本发明提供的将图像去 阴影方法应用于道路路面检测的方法流程框图，主要包括三大部分：
图像预处理。预处理的目的是取出无关区域和干扰信息，这里通过感兴趣区域选择去除掉无关区域，通过无阴影特征提取去除掉阴影的干扰；
路面提取。通过预处理后，现有的路面提取方法可以抵御阴影的干扰，不再在强阴影下失效。路面提取主要划分为两大步骤：图像分割和路面区域选择。通过图像分割，将道路图像划分成一个个区域；接着进行路面区域选择：对各个区域进行评估，选择最可能为路面的区域；
图像后处理。路面提取得到的路面候选区域并不完美，可能含有多余或者缺失的区域。通过图像形态学滤波，可以将多余的部分去除；道路上的车道标记可能会在区域中形成孔洞，采用孔洞填充算法对这些洞进行填补。完成后处理后，输出最终的道路检测结果。
实施例一：
在本发明实施例中，基于无阴影特征提取子的设计框架，提供一种适合于路面检测的无阴影特征提取子，其步骤如下：
离线相机标定。对应用场景的图像进行标定，即人工标出其中感兴趣的对象区域。标定工作可以使用交互软件完成。对于路面检测我们感兴趣的区域是路面，打开一张含阴影图片，圈出其中部分路面区域，如图1(a)所示，交互软件会将该区域内部的图像像素，如图1(b)所示，绘制到RGB空间如图1(c)，然后投影到GB空间，像素点云的分布可以由一个直线方程去拟合，软件对数据进行拟合得到直线在G轴上的截距b即为所需的相机参数，如图1(d)。
映射范围确定。得到相机参数b后，下一步是确定归一化的区间，即前面的m和n。以前面所选的道路图像为例，为简化运算，我们选用m＝1、n＝2作为归一化区间。此外，为了使得到的无阴影灰度图更加自然，我们做了反相处理(用1-C代替C)。最终得到的无阴影灰度图定义为：
Figure PCTCN2016109612-appb-000010
利用上述转换公式，即可将彩色道路图像转换为无阴影灰度图，并且实验结果表明，该公式不仅可应用于道路图像，对于大多数自然图像，该转换公式都能得到很好的去阴影效果。使用该方法得到的无阴影特征提取效果如图所示。
实施例二：
在本发明实施例中，提供一种道路路面检测方法。道路检测方法对一张驾驶员视角的道路图像进行处理，检测出其中路面所在的区域，如图3。所述道路路面检测的方法框架如图4 所示，具体包括以下步骤：
感兴趣区域(ROI,Region of interest)选择。该步骤将后续处理步骤限定于一个子区域内，以提高系统效率。对无关的区域进行处理不仅浪费时间，还可能引入噪声影响系统的性能。因此第一步是定义对于实施应用来说那些图像区域是没用的，哪些区域是有用的。对后续处理有用的区域称为感兴趣区域。另外，从简单角度考虑，一般选取矩形的感兴趣区域，按照矩形的ROI窗口划分后，直接得到一个子图像，方便了后续处理。对于道路检测问题，我们感兴趣的是路面部分，而在采集设备(如行车记录仪，车载摄像头)道路图像中，路面一般位于下半部分，而上半部分主要是天空，对道路检测是干扰信息。简单地，可以直接选取下半部分图像作为感兴趣区域，也可以根据实际应用场合做一定的调整，采用更加精确的ROI。
路面特征提取。一般采用颜色信息进行路面的分割，但由于路面可能出现大量阴影，采用颜色信息进行的分割可能失效。我们采用前面提出的无阴影特征提取方法，输入ROI内的彩色图像，输出一个无阴影的灰度图像。
图像滤波去噪。在原始图像中可能存在噪声，同时前面的路面特征提取也会引入噪声。该步骤通过对图像应用一组滤波器去除这些噪声。首先利用中值滤波去除其中的椒盐噪声，其次使用平滑滤波，过滤掉高频的噪声。
图像分割。采用图像分割算法对滤波后的图像进行分割，划分为一个个连通的区域。
路面区域选择。由于路面是ROI区域内最显著的区域，其占有面积也最大，通过连通分量分析即可提取出其中面积最大的路面区域。采用连通分量是在分割区域中筛选出路面区域的最简单的方法。也可以对区域特征分类选取最接近路面特征的区域，从而更加准确地提取出路面区域。
图像形态学滤波。连通分量分析得到的路面区域可能包含了一些无关的区域，对二值图像进行形态学滤波可以将这些无关区域去除掉。
孔洞填充。路面上的道路标记可能会在提取出的区域中形成若干孔洞，通过孔洞填充算法将这些孔洞填满，得到最终的路面检测结果。
需要注意的是，公布实施例的目的在于帮助进一步理解本发明，但是本领域的技术人员可以理解：在不脱离本发明及所附权利要求的精神和范围内，各种替换和修改都是可能的。因此，本发明不应局限于实施例所公开的内容，本发明要求保护的范围以权利要求书界定的范围为准。

Claims (10)

  1. 一种图像去阴影方法，包括无阴影特征分析提取过程、无阴影变换参数获取过程和无阴影特征成像过程；具体包括如下步骤：
    1)无阴影特征分析提取过程：对原始图像中同一材料在不同光照下的像素RGB值进行统计分析，或者通过物理成像模型找出含阴影图像中不变的特性，获得无阴影特征的表达式K＝f(R,G,B,…)，其中，K为无阴影特征矩阵；R、G、B分别对应原始图像的红绿蓝颜色的三个分量矩阵；…表示其他参数；f为转换表达式；
    2)无阴影变换参数获取过程：对步骤1)所述无阴影特征矩阵K＝f(R,G,B,…)，通过确定表达式中的参数，计算得到对应的无阴影特征矩阵K；
    3)无阴影特征成像过程：将所述无阴影矩阵转换为灰度图像，通过特征可视化得到无阴影特征灰度图像。
  2. 如权利要求1所述图像去阴影方法，其特征是，步骤1)利用RGB空间的线性关系定义所述无阴影特征的表达式；所述RGB空间的线性关系包括RGB每两个分量的线性关系或RGB三分量的线性组合的线性关系。
  3. 如权利要求1所述图像去阴影方法，其特征是，步骤1)选用B和G分量的线性关系表示用于路面检测的图像的无阴影特征，所述无阴影特征的表达式为式1：
    Figure PCTCN2016109612-appb-100001
    其中，b为待确定的其他参数，几何意义是拟合直线的截距；对于路面区域，K的取值为一个定值，K即为所求的无阴影特征矩阵。
  4. 如权利要求1所述图像去阴影方法，其特征是，步骤3)所述将无阴影特征矩阵转换为灰度图像，具体通过图像分割方法提取得到感兴趣区域材料的灰度值。
  5. 如权利要求4所述图像去阴影方法，其特征是，所述图像分割方法具体采用对无阴影特征矩阵进行归一化，使得矩阵各个元素的值分布在0和1之间，从而得到无阴影特征灰度图像。
  6. 如权利要求5所述图像去阴影方法，其特征是，通过求解最优划分区间的方法获得自然图像特征值最优区间的下界和上界m和n，再将区间(m,n)映射到区间(0，1)做归一化，得到无阴影灰度图C；所述最优划分区间通过式1～式3求解：
    分别计算得到一张或多张图像中的像素值对应的K的取值，将K取值的最大值和最小值分别记作Kmax和Kmin，统计直方图记作H：
    max:g(m,n)-c(m,n)         (式1)
    Figure PCTCN2016109612-appb-100002
    Figure PCTCN2016109612-appb-100003
    其中，m和n为所求最优区间的下界和上界；将(m,n)区间映射到(0，1)区间做归一化：
    Figure PCTCN2016109612-appb-100004
    得到最终的无阴影灰度图C。
  7. 权利要求1～6所述图像去阴影方法在机器视觉领域中的应用。
  8. 权利要求1～6所述图像去阴影方法在道路检测中的应用方法，其特征是，包括图像预处理过程、路面提取过程和图像后处理过程；首先通过无阴影特征分析提取过程进行感兴趣区域选择和特征提取，再进行图像滤波、分割和路面区域选择，最后进行图像形态学滤波和孔洞填充；具体包括如下步骤：
    81)进行图像预处理，取出无关区域和干扰信息；
    82)路面提取过程：包括图像分割和路面区域选择；通过所述图像分割，将道路图像划分成多个区域；通过所述路面区域选择对各个区域进行评估，选择得到最可能为路面的区域；
    83)图像后处理过程：通过图像形态学滤波，将多余的部分去除；采用孔洞填充算法对路面区域中的孔洞进行填补；
    完成后处理后，输出道路检测结果。
  9. 如权利要求8所述图像去阴影方法在道路检测中的应用方法，其特征是，步骤81)具体通过选择感兴趣区域去除无关区域；通过无阴影特征提取方法去除掉阴影的干扰，得到无阴影灰度图。
  10. 如权利要求9所述图像去阴影方法在道路检测中的应用方法，其特征是，所述无阴影灰度图为式5：
    Figure PCTCN2016109612-appb-100005
    式5中，C为无阴影灰度图；G、B分别对应图像的绿和蓝颜色的分量矩阵；b为相机参数，为拟合得到直线在G轴上的截距。
PCT/CN2016/109612 2016-08-01 2016-12-13 一种彩色图像去阴影方法和应用 WO2018023916A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/089,462 US10592754B2 (en) 2016-08-01 2016-12-13 Shadow removing method for color image and application

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610621447.XA CN106296666B (zh) 2016-08-01 2016-08-01 一种彩色图像去阴影方法和应用
CN201610621447.X 2016-08-01

Publications (1)

Publication Number Publication Date
WO2018023916A1 true WO2018023916A1 (zh) 2018-02-08

Family

ID=57663962

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/109612 WO2018023916A1 (zh) 2016-08-01 2016-12-13 一种彩色图像去阴影方法和应用

Country Status (3)

Country Link
US (1) US10592754B2 (zh)
CN (1) CN106296666B (zh)
WO (1) WO2018023916A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112598592A (zh) * 2020-12-24 2021-04-02 广东博智林机器人有限公司 图像阴影去除方法、装置、电子设备及存储介质
CN112837251A (zh) * 2020-12-31 2021-05-25 维沃移动通信有限公司 图像处理方法及装置
CN113223034A (zh) * 2021-05-17 2021-08-06 曲阜师范大学 一种道路边沿检测跟踪方法
CN114742724A (zh) * 2022-03-29 2022-07-12 中国科学技术大学 一种基于模型驱动的图像阴影去除方法

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109308711B (zh) * 2017-07-26 2021-11-05 富士通株式会社 目标检测方法、装置及图像处理设备
CN108629762B (zh) * 2018-03-28 2021-08-03 浙江工业大学 一种骨龄评测模型减少干扰特征的图像预处理方法及系统
CN111652806B (zh) * 2019-03-15 2023-05-12 上海铼锶信息技术有限公司 一种图像去阴影的方法和系统
CN110415296B (zh) * 2019-07-15 2023-03-17 上海合时智能科技有限公司 一种有阴影光照下矩形状电器件的定位方法
CN111161219B (zh) * 2019-12-10 2023-04-11 天津大学 一种适用于阴影环境的鲁棒单目视觉slam方法
CN111415357B (zh) * 2020-03-19 2023-04-07 长光卫星技术股份有限公司 一种基于彩色影像的便携式阴影提取方法
CN111488820B (zh) * 2020-04-08 2023-06-23 中国电力科学研究院有限公司 一种基于光影分离的电缆隧道工程智能巡检方法及系统
CN111968042B (zh) * 2020-07-08 2023-10-20 华南理工大学 一种数字文档的去阴影方法、系统、装置和存储介质
CN112561991B (zh) * 2020-11-20 2024-04-09 西安石油大学 基于surf特征提取和颜色分割的液位计图像识别方法
CN113240619B (zh) * 2021-01-26 2022-06-21 重庆师范大学 基于改进减法直方图的土壤图像阴影检测方法
CN113012059A (zh) * 2021-02-08 2021-06-22 瞬联软件科技（北京）有限公司 一种文字图像的阴影消除方法、装置及电子设备
DE102022206328B3 (de) 2022-04-19 2023-02-09 Continental Autonomous Mobility Germany GmbH Verfahren für ein Kamerasystem sowie Kamerasystem
WO2023202844A1 (de) 2022-04-19 2023-10-26 Continental Autonomous Mobility Germany GmbH Verfahren für ein kamerasystem sowie kamerasystem

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014130404A (ja) * 2012-12-28 2014-07-10 Aero Asahi Corp 路面画像生成システム並びに影除去装置、方法及びプログラム
CN104240207A (zh) * 2014-10-10 2014-12-24 深圳市开立科技有限公司 一种图像去阴影方法及装置
CN105488492A (zh) * 2015-12-25 2016-04-13 北京大学深圳研究生院 一种彩色图像预处理方法、道路识别方法及相关装置
CN105678285A (zh) * 2016-02-18 2016-06-15 北京大学深圳研究生院 一种自适应的道路鸟瞰图变换方法和道路车道检测方法
US20160180201A1 (en) * 2014-12-22 2016-06-23 International Business Machines Corporation Image processing

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7636455B2 (en) * 2002-06-04 2009-12-22 Raytheon Company Digital image edge detection and road network tracking method and system
US20130039574A1 (en) * 2011-08-09 2013-02-14 James P. McKay System and method for segmenting water, land and coastline from remote imagery
CN102509278A (zh) * 2011-10-31 2012-06-20 上海大学 基于Fisher判别准则的光照无关图去阴影方法
US9892318B2 (en) * 2015-12-22 2018-02-13 Here Global B.V. Method and apparatus for updating road map geometry based on received probe data

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014130404A (ja) * 2012-12-28 2014-07-10 Aero Asahi Corp 路面画像生成システム並びに影除去装置、方法及びプログラム
CN104240207A (zh) * 2014-10-10 2014-12-24 深圳市开立科技有限公司 一种图像去阴影方法及装置
US20160180201A1 (en) * 2014-12-22 2016-06-23 International Business Machines Corporation Image processing
CN105488492A (zh) * 2015-12-25 2016-04-13 北京大学深圳研究生院 一种彩色图像预处理方法、道路识别方法及相关装置
CN105678285A (zh) * 2016-02-18 2016-06-15 北京大学深圳研究生院 一种自适应的道路鸟瞰图变换方法和道路车道检测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
FINLAYSON, G.D.: "On the Removal of Shadows From Images", IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 31 January 2006 (2006-01-31), XP001523350, ISSN: 0162-8828 *
YING, ZHENQIANG: "A Novel Shadow-Free Feature Extractor for Real-Time Road Detection", PROCEEDINGS OF THE 2016 ACM ON MULTIMEDIA CONFERENCE, 19 October 2016 (2016-10-19), XP055603215 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112598592A (zh) * 2020-12-24 2021-04-02 广东博智林机器人有限公司 图像阴影去除方法、装置、电子设备及存储介质
CN112837251A (zh) * 2020-12-31 2021-05-25 维沃移动通信有限公司 图像处理方法及装置
CN112837251B (zh) * 2020-12-31 2024-04-02 维沃移动通信有限公司 图像处理方法及装置
CN113223034A (zh) * 2021-05-17 2021-08-06 曲阜师范大学 一种道路边沿检测跟踪方法
CN114742724A (zh) * 2022-03-29 2022-07-12 中国科学技术大学 一种基于模型驱动的图像阴影去除方法
CN114742724B (zh) * 2022-03-29 2024-04-02 中国科学技术大学 一种基于模型驱动的图像阴影去除方法

Also Published As

Publication number Publication date
CN106296666A (zh) 2017-01-04
US10592754B2 (en) 2020-03-17
CN106296666B (zh) 2019-03-01
US20190340446A1 (en) 2019-11-07

Similar Documents

Publication Publication Date Title
WO2018023916A1 (zh) 一种彩色图像去阴影方法和应用
CN110569704B (zh) 一种基于立体视觉的多策略自适应车道线检测方法
CN107330376B (zh) 一种车道线识别方法及系统
CN106651872B (zh) 基于Prewitt算子的路面裂缝识别方法及系统
CN106709436B (zh) 面向轨道交通全景监控的跨摄像头可疑行人目标跟踪系统
CN111079556A (zh) 一种多时相无人机视频图像变化区域检测及分类方法
CN103077521B (zh) 一种用于视频监控的感兴趣区域提取方法
CN103400150B (zh) 一种基于移动平台进行道路边缘识别的方法及装置
CN107292252B (zh) 一种自主学习的身份识别方法
WO2017041396A1 (zh) 一种车道线数据的处理方法、装置、存储介质及设备
CN111539273A (zh) 一种交通视频背景建模方法及系统
CN107909081B (zh) 一种深度学习中图像数据集的快速获取和快速标定方法
CN110738676A (zh) 一种结合RGBD数据的GrabCut自动分割算法
CN105740945A (zh) 一种基于视频分析的人群计数方法
US20130342694A1 (en) Method and system for use of intrinsic images in an automotive driver-vehicle-assistance device
CN106815583A (zh) 一种基于mser和swt相结合的夜间车辆车牌定位方法
CN103279755A (zh) 基于汽车底部阴影特征的快速车牌定位方法
CN106327464A (zh) 一种边缘检测方法
US9367759B2 (en) Cooperative vision-range sensors shade removal and illumination field correction
CN105913464A (zh) 一种基于视频的多人体目标在线测量方法
CN111161219B (zh) 一种适用于阴影环境的鲁棒单目视觉slam方法
Xiang et al. Research on target detection method based on HSV fusion Gaussian mixture model
CN113689365B (zh) 一种基于Azure Kinect的目标跟踪定位方法
CN106339709A (zh) 一种图像实时提取方法
CN113052118A (zh) 基于高速快球摄像机实现场景变换视频分析检测的方法、系统、装置、处理器及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16911511

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16911511

Country of ref document: EP

Kind code of ref document: A1