CN107154026B

CN107154026B - Method for eliminating road surface shadow based on self-adaptive brightness elevation model

Info

Publication number: CN107154026B
Application number: CN201710173817.2A
Authority: CN
Inventors: 裘国永; 李丽; 马卫飞
Original assignee: Shaanxi Normal University
Current assignee: Shaanxi Normal University
Priority date: 2017-03-22
Filing date: 2017-03-22
Publication date: 2020-12-18
Anticipated expiration: 2037-03-22
Also published as: CN107154026A

Abstract

The invention relates to a method for eliminating road surface shadows based on a self-adaptive brightness elevation model, which comprises the following steps of firstly, eliminating the influence of road surface cracks and road surface textures on the division of a subsequent shadow area by adopting morphological expansion operation and Gaussian smooth filtering; then, the division threshold values of the road surface image shadow region and the non-shadow region after Gaussian smoothing are solved by utilizing maximum entropy threshold value division, so that the self-adaptive determination of the division threshold values is realized. Finally, the elimination of the road surface shadow is realized based on an improved high-altitude area division model of the brightness and a brightness compensation method; the method for eliminating the road surface shadow based on the self-adaptive brightness elevation model not only effectively solves the problems that key parameters of a GSR method need to be manually set, and the defects of a method for dividing high-brightness areas and compensating the brightness exist, but also enhances the contrast ratio of road surface textures and road surface cracks to a certain extent, and achieves the purpose of eliminating the road surface shadow.

Description

Method for eliminating road surface shadow based on self-adaptive brightness elevation model

Technical Field

The invention belongs to the technical field of image processing, and particularly relates to a method for eliminating road surface shadows based on a self-adaptive brightness elevation model.

Background

In the period of high-speed development of highway traffic, with the increase of highway mileage and the continuous improvement of requirements of people on driving safety and comfort, an efficient pavement crack detection method has become the key point of research of scholars at home and abroad. However, the interfering objects such as buildings, trees, lamp posts and the like beside the road often form shadows on the road surface, which brings great challenges to the detection of the road surface cracks and the extraction of the characteristics of the road surface cracks. Therefore, elimination of shadows on the road surface before detection of the road surface cracks and extraction of the characteristics of the road surface cracks is crucial to maintenance and management of the road.

In order to eliminate shadows in images, researchers at home and abroad have conducted extensive and intensive research. Li et al propose a method for eliminating the self-adaptive non-local regularization shadow of the remote sensing aerial image on the basis of analyzing the shadow characteristics in the aerial image; wu et al studied the problem of shadow extraction in complex scenes; arbel et al assume that the brightness distribution of the penumbra area is an arc-shaped curved surface, and then realize the recovery of the penumbra area texture in the color image by a curved surface fitting method; liu et al, based on the original image, eliminates the shadows in the image by constructing a gradient region without shadows and with consistent texture, and using this as a template; mohan et al developed a shadow elimination editing software, set the intensity of brightness compensation by manually designating the boundary of the shadow region, and further realize the elimination of the shadow in the image; finlayson et al assume that the gradient of the transition boundary between the shadow region and the non-shadow region is zero, and eliminate the shadow on the basis of the gradient; avery et al studied the elimination of shadows in traffic images; ramamorthi et al studied drop shadows based on fourier theory; salamat et al are based on a probabilistic shadow map to eliminate shadows in real images. However, these shadow elimination methods do not completely solve the problems that the shadow area and the non-shadow area have inconsistent texture contrast after the shadow elimination, the brightness transition between the shadow area and the non-shadow area is unnatural, and the shadow cannot be automatically eliminated. In addition, the road shadow has the characteristics of huge penumbra area and extremely irregular shape, and if the shadow elimination method is directly used for carrying out shadow elimination on the road shadow image, the shadow elimination effect is quite unsatisfactory.

For the characteristics of road surface Shadow, Zou et al propose a Shadow elimination method (Geodesic Shadow-remove Algorithm) based on a brightness elevation model, which is abbreviated as GSR. The GSR method effectively solves the problems that a shadow area and a non-shadow area are difficult to define, and the contrast ratio of pavement cracks and pavement textures in the shadow area is not strong. However, the key parameters in the GSR method need to be set manually according to empirical values; in addition, the GSR method has a serious drawback in the partition model of the high-luminance region and the luminance compensation method.

Disclosure of Invention

The invention aims to solve the problems that the key parameters of the GSR method adopted by the existing method for eliminating the road shadow are selected by depending on experience, and the division model and the brightness compensation method of high-altitude areas have serious defects.

Therefore, the invention provides a method for eliminating road surface shadows based on a self-adaptive brightness elevation model, which comprises the following steps:

the method comprises the following steps: carrying out gray level operation on the collected road surface shadow image to obtain a gray level image of the road surface shadow;

step two: performing morphological expansion operation on the gray level image of the road surface shadow to eliminate the road surface crack;

step three: performing two-dimensional Gaussian smoothing on the result obtained by the morphological expansion operation in the step two;

step four: performing maximum entropy threshold segmentation on the road surface image after Gaussian smoothing, solving a segmentation threshold in the threshold segmentation process, taking the threshold as a segmentation threshold of a shadow region and a non-shadow region, and marking the segmentation threshold as metS, performing segmentation of the shadow region and the non-shadow region, and marking the shadow region as S and the non-shadow region as B;

step five: solving the standard deviation D of the brightness value of the pixel with the non-shaded area marked as B_BAnd average luminance I of non-shaded area_B'; standard deviation D of the brightness values of the pixels with the shaded area denoted S_sAnd average luminance I of non-shaded area_s’；

Step six: dividing the road surface image after Gaussian smoothing into high-brightness areas;

step seven: the whole image brightness compensation is performed by the formula (1),

I′_i,j＝α*I_i,j+λ (1)

wherein, I_i，jThe brightness value of each pixel point in the road surface image,

λ＝I_B’-α*I_S’。

performing morphological expansion operation on the grayscale image of the road shadow by the following formula

dilate(x,y)＝maxsrc(x+x’,y+y’)；(x’,y’)∈kernel (2)

Where, (x, y) is a pixel at a certain point of src in the road surface image, and (x ', y') is a certain element in kernel B.

The method for dividing the road surface image after Gaussian smoothing into equal-brightness equal-height areas is that the number of pixel points in each gray level and the number n of pixel points in each equal-brightness equal-height area are immediately divided after one line of the image is traversed in each gray level_gA comparison is made.

The invention has the beneficial effects that: the invention provides a road shadow elimination method based on a self-adaptive brightness elevation model, namely an SGRSR method, aiming at the defects that key parameters in a GSR method need to be manually set, and high-area division models such as brightness and the like and a brightness compensation formula are provided. Experimental results show that the method not only effectively solves the problems of the GSR method, but also enhances the contrast of the road surface texture and the road surface cracks to a certain extent, and achieves the purpose of eliminating the road surface shadow.

The present invention will be described in further detail below with reference to the accompanying drawings.

Drawings

Fig. 1 is a schematic diagram of a high-brightness region division algorithm of the GSR method.

Fig. 2 is a schematic diagram of an SGRSR method high-brightness region division algorithm.

Fig. 3a is an original drawing of a road surface shade.

FIG. 3b is the result of gauSmooth.

Fig. 3c is a luminance distribution diagram of sec.1.

Fig. 3d is the luminance distribution graph of sec.2.

Fig. 3e is the luminance distribution diagram of sec.3.

FIG. 3f is a diagram illustrating the results of maximum entropy threshold segmentation.

Fig. 4a is a shadow view of a large penumbra of the road surface.

Fig. 4b is a shadow view of an irregular road surface shape.

FIG. 4c is a crack pattern of low contrast in the shadow region of the pavement.

Fig. 4d is a distribution diagram of the number of pixels in each high brightness region of fig. 4 a.

Fig. 4e is a distribution diagram of the number of pixels in each high brightness region of fig. 4 b.

Fig. 5a is a schematic view of a road shadow region.

Fig. 5b is a graph of the results of GSR luminance compensation.

Fig. 6a is an original road surface image.

Fig. 6b is a diagram of the result of the closing operation.

Fig. 6c is a diagram of the result of the dilation operation.

Fig. 6d is a graph based on the closure elimination results.

Fig. 6e is a graph of the swelling-based elimination results.

Fig. 7a is an original road surface image two.

Fig. 7b is a graph two of the results of the closing operation.

Fig. 7c is a graph two of the results of the dilation operation.

Fig. 7d is a graph two based on the closure cancellation results.

Fig. 7e is a graph two of the swelling-based elimination results.

Fig. 8a is a high-luminance high-level division original road surface image.

Fig. 8b is a GSR luminance contour plot.

Fig. 8c is a plot of SGRSR brightness and the like.

Fig. 9a is a high-luminance high-division original road surface image two.

Fig. 9b is a GSR luminance contour plot two.

Fig. 9c is a second plot of SGRSR luminance contour.

Fig. 10a is an original road surface image.

Fig. 10b is a luminance compensation map of the GSR algorithm.

Fig. 10c is a luminance compensation diagram for the SGRSR algorithm.

Fig. 11a is an original road surface image map image.

Fig. 11b is a luminance compensation diagram two of the GSR algorithm.

Fig. 11c is a luminance compensation diagram two of the SGRSR algorithm.

Fig. 12a is an image of an experimental comparative original pavement.

Fig. 12b is a luminance compensation map of the experimental comparative GSR algorithm.

Fig. 12c is a graph of brightness compensation for the experimental comparison SGRSR algorithm.

Fig. 13a is an image of the original pavement image compared experimentally.

Fig. 13b is a luminance compensation map two of the experimental comparison GSR algorithm.

Fig. 13c is a luminance compensation graph two of the experimental comparison SGRSR algorithm.

Fig. 14a is a three-dimensional image of an experimental comparative original pavement.

Fig. 14b is a luminance compensation map three of the experimental comparison GSR algorithm.

Fig. 14c is a graph of brightness compensation for the experimental comparison SGRSR algorithm three.

Fig. 15a is an image of the original pavement image for experimental comparison.

Fig. 15b is a luminance compensation map four of the experimental comparison GSR algorithm.

Fig. 15c is a graph of brightness compensation for the experimental comparison SGRSR algorithm four.

FIG. 16a is a schematic diagram of a simulated dilation kernel.

Fig. 16b is a schematic diagram of a simulated pavement crack image.

FIG. 16c is a graph showing the result of the simulation of the image after the first sliding of the dilation operation.

Detailed Description

To further explain the technical means and effects of the present invention adopted to achieve the intended purpose, the following detailed description of the embodiments, structural features and effects of the present invention will be made with reference to the accompanying drawings and examples.

Example 1

For better contrast with the existing shadow elimination method (GSR) based on brightness elevation, the present embodiment first describes how the road surface shadow elimination is performed by the GSR method.

The steps can be roughly divided into the following four steps:

(1) mmClose-form closure operation. And processing the original pavement image by adopting morphological closed operation to remove the cracks of the pavement. The purpose of this operation is to eliminate the influence of the crack on the subsequent shadow area division.

(2) gauSmooth-Gaussian smoothing. And performing two-dimensional Gaussian smoothing on the result obtained in the last step. The purpose of this step of operation is to smooth the road surface texture and eliminate the influence of the road surface texture on the subsequent shadow area division.

(3) And dividing geoLevel-brightness equal-height areas. First, N-1 thresholds, 0, are calculated<＝K1<＝K2<＝...<＝KN-1<255, the image obtained in the previous step is divided into regions { Gi | I ═ 1., L., N } of different brightness heights, so that the region Gi includes a brightness value I ∈ (K ∈ N) }_i-1,K_i]Where K0-1 and KN-255, see Algorithm 1 for details of the method. In order to make the method have general applicability, Algorithm 1 keeps the number of pixels in each high-brightness area consistent; next, L lower luminance levels S ═ { Si ═ Gi | I ═ 1, 2., L } are selected as the shaded regions, and N-L higher luminance levels B ═ Gi | I ═ L +1, L + 2.,. N } are selected as the unshaded regions, where L takes the empirical value (7/8) × N. The procedure is shown in FIG. 1.

(4) illumComp-illumination compensation with texture equalization. In the original road surface image, the brightness compensation method corresponding to the formula (1) is applied to perform brightness compensation on the shadow region Si of each level.

Wherein α ═ D_B/D_S，λ＝I_B’-α*I_S’，D_SAnd D_BStandard deviations, I, representing pixel intensity values of shadow and non-shadow regions, respectively_B' and I_S' denotes an average value of pixel luminance of a non-shaded area and a shaded area, respectively.

Example 2

The embodiment provides a Road Shadow eliminating method (Self-adaptive geochemical Road-Shadow-remove Algorithm) based on a Self-adaptive brightness elevation model, which comprises the following steps:

(1) GrayImg-graying of image calculation. The collected RGB road surface shadow image is converted into a gray level image through the gray level operation of the image, and the main purpose is as follows: the required gray scale image of the road surface shadow image is prepared for the subsequent processing of the shadow elimination method.

(2) mmDilate-morphological dilation operation. After the three-channel road surface image is grayed, the three-channel road surface image is converted into a single-channel grayscale image. In the grayscale image of the road surface crack, the grayscale level of the road surface crack is low and is mostly closer to black, and the characteristics of the road surface shadow are shown in fig. 4a, 4b and 4 c. Therefore, the purpose of eliminating the road surface cracks can be achieved by using the expansion operation in morphology.

The expansion operation is a local maximum. Mathematically, the dilation operation is the convolution of an image (or a portion of an image, referred to as a) with a kernel (referred to as B). The maximum value of the pixel points of the region covered by the kernel B is calculated and attached to the specified pixel of the reference point. This will cause the high brightness regions in the image to grow gradually, as shown in fig. 6a and fig. 6c, where fig. 6a is the original region containing pavement cracks and pavement shadows; FIG. 6c is the result of FIG. 6a after a dilation operation; after the dilation operation of fig. 6a, the black crack in fig. 6a is covered by the growing high luminance region due to the gradual growth of the high luminance region in the image. The mathematical expression for the expansion operation is as in equation (2).

dilate(x,y)＝maxsrc(x+x’,y+y’)；(x’,y’)∈kernel (2)

Where (x, y) is a pixel at a certain point of src in the road surface image, and (x ', y') is a certain element in kernel B.

The nucleus can be generally four kinds of methods, namely rectangular nucleus, cross nucleus, elliptical nucleus and any nucleus which can be customized, and the following common expansion nuclei are exemplified as follows:

say a cross kernel of 3 x 3 pixels size, we compute the convolution of this kernel and the pixels of the image under the kernel coverage, and then take the maximum value, by sliding this cross kernel from the (0, 0) coordinates of the image, each time. Thus, after the image is slid, the functions of expanding the white area and reducing the black area can be achieved; in the case of a road surface crack image, it is the crack of the road surface that is eliminated by this swelling action.

And sliding the kernel on the image, calculating the product of the kernel and a pixel point corresponding to the image area covered by the kernel once every sliding, and taking the maximum value of the area as the value of a reference point, so that the brightness area in the image is increased. The specific calculation process is shown in the following graph, for example, now an expansion kernel is simulated by using fig. 16a, fig. 16b simulates a pavement crack image, 9 in the pavement crack image represents the brightness of a high-brightness area or an area without cracks of the pavement, a black band composed of squares with 0 represents the pavement crack, and 0 represents the brightness of an area with cracks (the brightness of cracks is darker); now, the kernel represented by fig. 16a is subjected to coverage sliding from the upper left corner of fig. 16b, each time sliding is performed, the product of corresponding pixels of the kernel and the area covered by the kernel is calculated, then the maximum value of the calculation result of the area covered by the kernel is taken as the pixel value of the reference point (namely, the gray area), the calculation result of the first sliding is shown in fig. 16c, we can clearly see that the pixel value of the reference point is changed into the gray value of the normal road surface, then we continuously slide the kernel on the image, so that after one sliding is completed, the luminance area of the image is increased, the road surface crack in the image is disappeared, and the above is the specific process of the expansion operation.

Based on the above analysis, the SGRSR method improves the morphological closing operation in the GSR method to the morphological dilation operation. In addition, the dilation operation in the SGRSR method is greatly superior to the morphological closure operation in the GSR on the method performance; moreover, the shadow elimination method based on the expansion operation can better realize the elimination of the road shadow.

(3) gauSmooth-Gaussian smoothing. Two-dimensional Gaussian smoothing is carried out on the result obtained by the morphological dilation operation in the previous step, and the main purposes are as follows: and smoothing the road texture to eliminate the influence of the road texture on the subsequent shadow area division.

Gaussian smoothing to reduce the degree of blurring in image smoothing processing and to obtain a more natural smoothing effect, the idea is to appropriately increase the weight of the center point of the template, and as the distance from the center point increases, the weight decreases rapidly, thereby ensuring that the center point appears closer to the point closer to it. The 3 x 3 gauss template W commonly used is shown in formula (2). The gaussian template is just a discretization representation of a continuous two-dimensional gaussian function, so that the gaussian template with any size can be obtained by establishing a matrix of (2k +1) × (2k +1), wherein the element value of the point (i, j) position in the road image can be determined by the gaussian function formula (3).

(4) metSegment-maximum entropy threshold segmentation. The method comprises the steps of carrying out maximum entropy threshold segmentation on a road surface image after Gaussian smoothing, solving a segmentation threshold in the threshold segmentation process, taking the threshold as a segmentation threshold of a shadow region and a non-shadow region, and recording the threshold as metS, so that the problem that the segmentation threshold L in the GSR method needs to be manually set according to an empirical value is solved, and the self-adaption determination of the segmentation threshold is realized. Fig. 3f is the result of maximum entropy threshold segmentation of the road shadow image after gaussian smoothing. In addition, the positions of a shadow area and a non-shadow area in the road surface image are determined through maximum entropy threshold segmentation. The shaded area is denoted as S and the unshaded area is denoted as B.

(5) BriightParameter-solving key parameter D of non-shadow area according to the non-shadow area of road surface determined in the previous step_BAnd I_B’。D_BIs the standard deviation of the luminance values of the pixels in the unshaded region, I_B' represents the average luminance of the non-shaded area. The purpose of this operation is to prepare for the subsequent full image brightness compensation. In the brightness compensation formula (1) of the GSR method, where α ═ D_B/D_S，λ＝I_B’-α*I_S’，D_SAnd D_BStandard deviations, I, representing pixel intensity values of shadow and non-shadow regions, respectively_B'and IS' denote the average values of the pixel brightness of the non-shaded area and the shaded area, respectively. By introducing the parameter alpha, the variance of the shadow area can be improved to the level of the non-shadow area; and because the size of the image pixel variance generally reflects the strength of the image contrast, the texture details of the road surface image after the shadow is eliminated can be kept consistent by the brightness compensation formula of the GSR method.

However, after intensive research on the brightness compensation formula in the GSR, it is found that the key parameter α in the brightness compensation formula has a significant drawback. As can be known from the calculation method of α in the GSR method, when Ds tends to be infinitesimal, that is, when the contrast of the shadow area in the image is extremely weak, the value of α tends to be infinite, so that the brightness compensation formula in the GSR method fails. Aiming at the problem, the calculation mode of alpha is improved by the SGRSR method, and the calculation formula after the improvement is as follows:

wherein the meanings of the parameters alpha, DS and DB are consistent with those in the GSR algorithm.

(6) igeoLevel-improved brightnessAnd (5) dividing the height equal region into models. The method for dividing the high-brightness areas in the GSR method, namely, the Algorithm 1 is carefully analyzed, so that the Algorithm 1 cannot ensure that the number of pixel points in each high-brightness area is generally consistent; this defect of Algorithm 1 causes an extremely unnatural brightness transition in a road image after the shadow is removed. Fig. 4d and 4e are histograms showing the distribution of the number of pixels in the high brightness regions after the two road images of fig. 4b and 4c are subjected to high brightness region division by Algorithm 1. Wherein the abscissa of the histogram represents the brightness level and the ordinate represents the number of pixel points in this brightness level region; fig. 4d is a distribution diagram of the number of pixels in each high-brightness region in fig. 4b, and fig. 4e is a distribution diagram of the number of pixels in each high-brightness region in fig. 4 c. The defects of Algorithm 1 are clearly shown by the two histograms of FIG. 4d and FIG. 4 e. Aiming at the problem, the SGRSR method improves the Algorithm 1 in the GSR, and specifically, the method is to compare the number of pixels in the gray level with ng after traversing each gray level in Algorithm 1 once through the image instead of comparing the number of pixels in the gray level with ng immediately after traversing one line of the image in each gray level, and the improved method is as in Algorithm 2. According to Algorithm 2, the road surface image subjected to gaussian smoothing is divided into N different high-brightness regions { Gi | I ═ 1., L., N }, so that the region Gi contains a brightness value I ∈ (K ∈ K })_i-1,K_i]Where K0-1 and KN-255, the algorithm is shown in fig. 2.

(7) AllillumComp-full image luminance compensation.

As can be known from the brightness compensation formula (1) in the GSR method, the brightness compensation in the GSR method is only performed for the shadow region, and when a pixel belongs to the non-shadow region, the brightness of the pixel is not compensated. However, the brightness compensation formula in the GSR method cannot accurately calculate the brightness difference between the brightness of each pixel point in the shadow region and the brightness of the pixel point under the same sunlight intensity; therefore, when the road surface shadow is eliminated, if only the brightness compensation is performed on the pixel points of the shadow region, there may be a problem that the brightness transition between the shadow region and the non-shadow region of the road surface image after the shadow elimination is unnatural, and the specific effect can be referred to fig. 5a-5b, where fig. 5a is a gray scale image of the road surface crack shadow, and the right image is a road surface image with uneven brightness distribution and unnatural transition. The road surface image obtained by performing the shading removal by the GSR method on fig. 5a is shown in fig. 5b, and it can be clearly found that the right image has an unnatural luminance transition. In response to this problem, the SGRSR method improves the brightness compensation formula in the GSR method, and the brightness compensation formula after the improvement is shown in formula (5).

I′_i,j＝α*I_i,j+λ (5)

According to the improved brightness compensation formula (5), wherein I_i，jThe significance and the calculation mode of other parameters are the same as those of the GSR method for the brightness value of each pixel point in the road surface image. In the gray level image of the original road surface image, according to the high-brightness areas divided in the previous step, the variance D of the pixel brightness of each high-brightness area is firstly calculated_S' and average value of pixel luminance I_S', and alpha values for each of the luminance contour regions; finally, the brightness compensation is performed on the full image according to the brightness compensation formula (5).

Compared with the brightness elevation based shadow elimination method (GSR) described in the embodiment, the adaptive brightness elevation model based road shadow elimination method (SGRSR) described in this embodiment has the following advantages:

(1) in order to eliminate the pavement cracks, a shadow elimination method based on expansion operation is used, so that the pavement shadow can be eliminated better; naturally, the purpose of eliminating the road surface cracks can be achieved by using the morphological closing operation, but the morphological closing operation is to perform the expansion operation first and then perform the corrosion operation, which not only invisibly increases the time complexity of the method, but also makes the gray intensity change at the junction of the shadow area and the non-shadow area more severe, which will cause the unnatural transition of the brightness of the shadow area and the non-shadow area in the road surface image after the shadow elimination.

(2) In the use of dividing the shadow area and the non-shadow area, the maximum entropy threshold value is adopted for separation, so that the mode that the dividing threshold value L of the shadow area and the non-shadow area in the GSR method needs to be manually set according to experience is improved. Especially, the shadow elimination for the massive road surface images is very unscientific. In order to solve the problem, the document deeply analyzes the road surface shadow image after Gaussian smoothing. As shown in fig. 3b, the present application takes three section lines sec.1, sec.2, and sec.3 in the road surface image after gaussian smoothing, and the three section lines pass through the shaded area and the non-shaded area respectively. Fig. 3c, 3d, and 3e are luminance distribution histograms of the hatching lines sec.1, sec.2, and sec.3, respectively, and it is understood from the observation of the three hatching luminance histograms and fig. 3a and 3b that the luminance value of the gray scale changes drastically in the boundary region between the shadow region and the non-shadow region.

Fig. 3f is the result of maximum entropy threshold segmentation of the road shadow image after gaussian smoothing. In addition, the positions of a shadow area and a non-shadow area in the road surface image are determined through maximum entropy threshold segmentation. The shaded area is denoted as S and the unshaded area is denoted as B.

(3) An improved brightness equal-height area division model sets the number of pixel points in each brightness equal-height area as n_gIn the SGRSR method, the number of pixels in each gray level and n are determined after each gray level in the GSR method traverses the image once_gComparing, namely, immediately traversing one line of the image in each gray scale, and then comparing the number of pixel points in the gray scale with n_gCompared with the prior art, the method solves the problem that the luminance transition of the road image after the shadow is eliminated is extremely unnatural because the GSR method cannot ensure that the number of pixel points in each high-luminance region is generally consistent.

(4) Improving the parameter alpha of the brightness compensation formula by alpha-D_B/D_SIs improved by

Thereby correcting the GSR method when Ds tends to be infinitesimal smallThat is, when the contrast of the shadow area in the image is very weak, the value of α will tend to be infinite, thereby disabling the original brightness compensation formula.

Example 3

In order to verify the effectiveness of the method for eliminating the road surface shadow based on the adaptive brightness elevation model, four groups of comparison experiments are designed respectively and are used for carrying out quantitative and qualitative analysis on key steps in the SGRSR method. The program of the method is developed based on VC, OpenCv and Matlab, the running environment of the program is Windows10, the CPU is 3.3GHz, and the memory is 8G.

Four sets of comparative experiments were designed: the first set of experiments was used to test the improvement of the SGRSR method on morphological closing operations in the GSR method. The results of the experiment are shown in Table 1 and FIGS. 6a-6e, 7a-7 e.

Table 1 comparison of performances of pavement crack elimination method by SGRSR expansion operation and GSR closing operation

The experiments show that the morphological dilation operation in the SGRSR method can also achieve the aim of eliminating the road surface cracks through the morphological closure operation in the GSR method; meanwhile, the expansion operation in the SGRSR method is greatly superior to the morphological closed operation in the GSR in the method performance; moreover, the shadow elimination method based on the expansion operation can better realize the elimination of the road shadow.

The second set of experiments was used to test the improvement of the SGRSR method on the high-intensity region partitioning model in the GSR method. Fig. 8a to 8c and fig. 9a to 9c show experimental results, where fig. 8b and 9b show distribution histograms of the number of pixels in each high-luminance region after being divided by the high-luminance region division model in the GSR method, and fig. 8c and 9c show distribution histograms of the number of pixels in each high-luminance region after being divided by the high-luminance region division model improved in the SGRSR method. As is clear from fig. 8a to 8c and fig. 9a to 9c, the improved partition model in the SGRSR method can make the number of pixels in each high-brightness region substantially consistent, which is beneficial to enhancing the applicability of the SGRSR method to other shadow elimination, and makes the brightness of the road image after the shadow elimination more natural.

A third set of experiments was used to test the improvement of the SGRSR method on the brightness compensation formula in the GSR method. The results of the experiment are shown in FIGS. 10a to 10c and FIGS. 11a to 11 c. As can be observed from fig. 10a to 10c and fig. 11a to 11c, the problem of uneven brightness distribution of the road surface image occurs after the shadow elimination is performed based on the brightness compensation formula in the GSR, and the defect of the brightness compensation formula in the GSR method can be well solved based on the brightness compensation formula after the SGRSR improvement.

The fourth set of experiments was used to test the road surface shadow removal capability of both the SGRSR method and the GSR method. The final results of the experiment are shown in FIGS. 12a-12c, FIGS. 13a-13c, FIGS. 14a-14c, and FIGS. 15a-15 c. Through the comparison of all the images in fig. 12a-12c, fig. 13a-13c, fig. 14a-14c, and fig. 15a-15c, it can be clearly observed that, after the road surface shadow is eliminated based on the GSR method, the road surface image not only has the unnatural brightness transition between the shadow area and the non-shadow area, but also has the problem of uneven brightness distribution of the whole road surface image; the SGRSR method provided by the invention not only can well solve the problems of the GSR method, but also can enhance the contrast ratio of the road texture and the road crack, and the phenomena show that the SGRSR method provided by the invention is far superior to the GSR method, and can meet the requirement of eliminating the road shadow in the actual engineering.

The above description is a detailed description of the present embodiment with reference to specific preferred embodiments, and it should not be construed that the specific implementations of the present embodiment are limited to these descriptions. For those skilled in the art to which the embodiment belongs, several simple deductions or substitutions may be made without departing from the concept of the embodiment, and all of them should be considered as belonging to the protection scope of the embodiment.

Claims

1. A method for eliminating road surface shadows based on an adaptive brightness elevation model is characterized by comprising the following steps:

I'_i,j＝α*I_i,j+λ (1)

λ＝I_B’-α*I_S’。

2. the method for eliminating the road surface shadow based on the adaptive brightness elevation model according to claim 1, wherein the method comprises the following steps: performing morphological expansion operation on the grayscale image of the road shadow by the following formula

dilate(x,y)＝max src(x+x',y+y')；(x',y')∈ker nel

(2)

Wherein, (x, y) represents the coordinate of a certain pixel point in the original image, and src (x, y) represents the gray value of the pixel point at the coordinate (x, y); (x ', y') represents the coordinates of the pixel points in kernel B; max src (x + x ', y + y') represents that the original image coordinate point (x, y) is used as an anchor point, then the coordinates of the central point of the kernel B kernel are overlapped with the (x, y) point, the maximum value of the gray value of the pixel of the area under the coverage of the kernel B is obtained, and then the gray value at the original image coordinate point (x, y) is replaced by the maximum value.

3. The method for eliminating the road surface shadow based on the adaptive brightness elevation model according to claim 1, wherein the method comprises the following steps: the method for dividing the road surface image after Gaussian smoothing into equal-brightness equal-height areas is that the number of pixel points in each gray level and the number n of pixel points in each equal-brightness equal-height area are immediately divided after one line of the image is traversed in each gray level_gA comparison is made.