Road sign recognition method

By establishing a color difference characterization model and performing binarization processing, combining the color and shape characteristics of the connecting area, identifying road signs, solving the difficulty of extracting sign areas in the prior art and improving the recognition accuracy.

CN114511770BActive Publication Date: 2024-11-29WUHAN WUDA ZOYON SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202111573554.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-21
Publication Date
2024-11-29
Estimated Expiration
2041-12-21

AI Technical Summary

Technical Problem

In the prior art, the color difference of the sign signs is large due to background lighting, and the shape characteristics difference is caused by factors such as the deformation of the sign signs, camera acquisition angle, background occlusion, etc., making it difficult to accurately extract the sign signs based on color space transformation. Suspected target area.

Method used

By establishing a color difference characterization model, binarization and denoising the detection image, determining the color characteristics and shape characteristics confidence of the connecting area, and identifying the connecting area of ​​the road sign.

Benefits of technology

The accuracy of extracting the marked sign area based on the fixed threshold segmentation method based on color space conversion is improved, and target misjudgment caused by factors such as deformation, angle and occlusion is avoided.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114511770B_ABST
    Figure CN114511770B_ABST
Patent Text Reader

Abstract

The present invention provides a method for identifying road signs and signboards, comprising: establishing a color difference characterization model for each pixel point in an image to be detected, so as to obtain a characterization image according to the color difference characterization model; binarizing the characterization image to obtain a binarized image, and denoising the binarized image to obtain a first suspected target area of ​​the binarized image, wherein the binarized image includes M valid pixel points, and the valid pixel points are pixel points whose color difference characterization values ​​are higher than a first preset threshold value; based on each of the valid pixel points and the first suspected target area, determining a second suspected target area of ​​the binarized image, wherein the second suspected target area includes N connected areas, and M and N are both positive integers; based on the color feature confidence and shape feature confidence of each of the connected areas, determining a connected area corresponding to the road sign and signboard.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method for recognizing road signs and labels. Background Art

[0002] Road sign detection is not only an important part of the intelligent transportation system, but also can provide timely information on road sign damage, missing and color degradation, which is of great significance to reducing traffic accidents and increasing safe driving.

[0003] At present, the method for detecting road signs in related technologies mainly relies on color space conversion, using a fixed threshold to extract suspected target areas that meet the color characteristics of the signs, and then using a shape feature-based method to determine the sign area in an image based on the specific shape of the signs.

[0004] However, due to the large differences in the colors of signs caused by background lighting, it is difficult to extract the suspected target area of ​​signs using the existing color threshold segmentation method based on color space transformation. In addition, due to the differences in shape features caused by factors such as sign deformation, camera acquisition angle, and background occlusion, it is not conducive to the final confirmation of the sign target. Summary of the invention

[0005] The present invention provides a method for recognizing road signs, which is used to solve the problem in the prior art that the color differences of signs are large due to background lighting, and the shape characteristics are different due to factors such as sign deformation, camera acquisition angle, background occlusion, etc., which makes it difficult to extract suspected target areas of signs using the existing color threshold segmentation method based on color space transformation.

[0006] The present invention provides a road sign recognition method, comprising:

[0007] Establishing a color difference representation model for each pixel in the image to be detected, so as to obtain a representation image according to the color difference representation model;

[0008] Binarizing the representation image to obtain a binary image, and performing denoising on the binary image to obtain a first suspected target area of ​​the binary image, wherein the binary image includes M valid pixel points, and the valid pixel points are pixel points whose color difference representation values ​​are higher than a first preset threshold value;

[0009] Determine a second suspected target region of the binary image based on each of the valid pixel points and the first suspected target region, wherein the second suspected target region includes N connected regions, and M and N are both positive integers;

[0010] Based on the color feature confidence and shape feature confidence of each of the connected areas, a connected area corresponding to the road sign is determined.

[0011] According to a road sign recognition method provided by the present invention, the method of establishing a color difference representation model for each pixel point in the image to be detected to obtain a representation image according to the color difference representation model includes:

[0012] Based on the RGB color information of each pixel point, the maximum value of the RGB color information difference, the median value of the RGB color information and the median value of the RGB color information difference of each pixel point are obtained;

[0013] Based on the maximum value of the RGB color information difference of the pixel point, the median value of the RGB color information and the median value of the RGB color information difference, a color difference representation value of each pixel point is obtained;

[0014] Based on each of the color difference characterization values, a characterization image is obtained.

[0015] According to a road sign recognition method provided by the present invention, the binarization process is performed on the representation image to obtain a binarized image, and the binarization image is denoised to obtain a first suspected target area of ​​the binarized image, including:

[0016] Pixels whose color difference representation values ​​are greater than the first preset threshold are regarded as valid pixels, and pixels whose color difference representation values ​​are less than or equal to the first preset threshold are regarded as other pixels;

[0017] A binary image is obtained based on the effective pixel points and other pixel points, and a denoising process is performed on the binary image to obtain a first suspected target area of ​​the binary image.

[0018] According to a road sign recognition method provided by the present invention, the second suspected target area of ​​the binary image is determined based on each of the effective pixel points and the first suspected target area, including:

[0019] Based on the valid pixel point, determine a neighborhood area corresponding to the pixel point;

[0020] Compare the candidate pixels in the neighborhood area with the valid pixels, determine candidate pixels among the candidate pixels, and use the candidate pixels as valid pixels, wherein the candidate pixels in the neighborhood area are pixels having color difference characterization features or RGB color information features of the valid pixels;

[0021] All the valid pixels are traversed to determine a second suspected target area of ​​the binary image.

[0022] According to a road sign recognition method provided by the present invention, the method of determining the connected area corresponding to the road sign based on the color feature confidence and shape feature confidence of each connected area includes:

[0023] Determine the color feature confidence of each of the connected regions based on the R channel grayscale value, the G channel grayscale value, and the B channel grayscale value of each pixel in the connected region;

[0024] After smoothing the original outer contours of each of the connected regions, a fitting process is performed to obtain the target outer contours of each of the connected regions;

[0025] Determine the shape feature confidence of each of the connected regions based on the similarity between the connected regions corresponding to the original outer contours of the connected regions and the connected regions corresponding to the target outer contours;

[0026] Based on the shape feature confidence and the color feature confidence of each of the connected regions, obtaining the target confidence of each of the connected regions;

[0027] Determining the connected area corresponding to the road sign based on the target confidence of each of the connected areas;

[0028] The step of obtaining the target confidence of each of the connected regions is specifically as follows:

[0029]

[0030] Among them, C1 is the color feature confidence, C2 is the shape feature confidence, ω1 and ω2 are the feature allocation weights for the color feature confidence and the shape feature confidence, respectively.

[0031] According to a road sign recognition method provided by the present invention, the color feature confidence of each connected area is determined based on the R channel grayscale value, G channel grayscale value and B channel grayscale value of each pixel point in the connected area, including:

[0032] The grayscale values ​​of the R channel, the G channel and the B channel of each pixel are sorted by size to obtain the grayscale value sorting information of each pixel;

[0033] The gray value sorting information of each pixel point specifically includes:

[0034]

[0035] The color difference representation value corresponding to each of the gray value sorting information is obtained, and the color feature confidence of each of the connected regions is determined based on each of the color difference representation values.

[0036] A road sign recognition method and device provided by the present invention performs color difference characterization processing on each pixel point in the image to be detected, so as to obtain a characterization image that can more effectively extract targets with color differences in natural scenes, thereby improving the accuracy of extracting suspected sign areas by a fixed threshold segmentation method based on color space conversion, and further determining the sign area through the color feature information and shape feature information of each connected area, thereby avoiding the problem of misjudgment of the final sign target due to sign deformation, camera acquisition angle, background occlusion and algorithm fitting method. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A schematic diagram of a road sign recognition method provided in an embodiment of the present application; DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Figure 1 A schematic diagram of a road sign recognition method provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, including:

[0041] Step 110, establishing a color difference characterization model for each pixel in the image to be detected, so as to obtain a characterization image according to the color difference characterization model;

[0042] Specifically, the image to be detected described in the embodiments of the present application may be an RGB image captured by an image acquisition device and may contain road signs.

[0043] In the embodiment of the present application, the color difference characterization model can be specifically based on the color difference of each pixel in the image to be detected, and the color difference characterization model is established, and the data of the RGB space is mapped to a space characterized by color difference. The characterization space first obtains the maximum value of the RGB color information difference, the median value of the RGB color information, and the median value of the RGB color information difference of each pixel, and then establishes a color difference characterization model based on the color information difference value to obtain the color difference characterization value of each pixel, thereby obtaining the characterization image corresponding to the image to be detected.

[0044] 120, binarizing the representation image to obtain a binarized image, and denoising the binarized image to obtain a first suspected target region of the binarized image, wherein the binarized image includes M valid pixel points, and the valid pixel points are pixel points whose color difference representation values ​​are higher than a first preset threshold value;

[0045] Specifically, in the embodiment of the present application, the representation image is binarized according to the representation value of each pixel, that is, the valid pixels and candidate pixels in each pixel are determined, so that the grayscale value of the valid pixel can be set to 255, and the grayscale value of the candidate pixel can be set to 0, or the grayscale value of the valid pixel can be set to 0, and the grayscale value of the candidate pixel can be set to 255, so as to obtain the binary image corresponding to the representation image.

[0046] More specifically, in the embodiment of the present application, after the binary image is obtained, it will be further subjected to denoising processing, so as to ensure the accuracy of the first suspected target area finally determined.

[0047] The first suspected target area described in the embodiment of the present application may specifically be an area including road signs.

[0048] More specifically, the first preset threshold described in the embodiment of the present application is a numerical value that can be pre-set, or it can be a numerical value adaptively obtained based on the current representation image, and the pixel point whose color difference representation value is higher than the first preset threshold indicates that the pixel point may be a pixel point belonging to a road sign, and the embodiment of the present application regards it as a valid pixel point.

[0049] Step 130, determining a second suspected target region of the binary image based on each of the valid pixels and the first suspected target region, wherein the second suspected target region includes N connected regions, and M and N are both positive integers;

[0050] In the embodiment of the present application, valid pixel points in the first suspected target area are used as seed points, and other points in the neighborhood of the seed point are marked by using the same features as the current seed point, so as to obtain the second suspected target area. The same features can specifically include the distribution of the R, G, and B three-channel values ​​or the characterization value based on color difference.

[0051] Step 140: Determine the connected area corresponding to the road sign based on the color feature confidence and shape feature confidence of each of the connected areas.

[0052] Specifically, the color feature information described in the embodiments of the present application may be a method for establishing a target area classification method using the color features of the sign, and then using the target area classification method to calculate a confidence level based on the color features.

[0053] The shape feature information described in the embodiment of the present application can specifically calculate the similarity between the connected area and the outer contour corresponding to the connected area after smoothing the original outer contour of each connected area and then fitting it, so as to determine the confidence of the shape features of each connected area.

[0054] After obtaining the color feature information and shape feature information of each connected area, the target confidence of each connected area can be further determined, and based on the comparison between the target confidence and the second preset threshold, it is judged whether the connected area belongs to the connected area corresponding to the road sign.

[0055] In an embodiment of the present application, by performing color difference characterization processing on each pixel point in the image to be detected, a characterization image is obtained that can more effectively extract targets with color differences in natural scenes, thereby improving the accuracy of extracting suspected sign areas by a fixed threshold segmentation method based on color space conversion. At the same time, the sign area is further determined by the color feature information and shape feature information of each connected area, thereby avoiding the problem of misjudgment of the final sign target due to sign deformation, camera acquisition angle, background occlusion and algorithm fitting method.

[0056] Optionally, establishing a color difference characterization model for each pixel in the image to be detected to obtain a characterization image according to the color difference characterization model includes:

[0057] Based on the RGB color information of each pixel point, the maximum value of the RGB color information difference, the median value of the RGB color information and the median value of the RGB color information difference of each pixel point are obtained;

[0058] Based on the maximum value of the RGB color information difference of the pixel point, the median value of the RGB color information and the median value of the RGB color information difference, a color difference representation model is established to obtain a color difference representation value of each pixel point;

[0059] Based on each of the color difference representation values, obtaining a representation image;

[0060] The color difference characterization model is specifically:

[0061]

[0062] T ij = mid{|R ij -G ij |,|R ij -B ij |,|G ij -B ij |}

[0063] Among them, R ij , G ij , B ij The grayscale values ​​of the three channels R, G, and B corresponding to the pixel point in the i-th row and j-th column respectively. max is the maximum value in the set, min is the minimum value in the set, and mid is the median value in the set.

[0064] Specifically, the maximum value of the RGB color information difference of each pixel point described in the embodiment of the present application can be max{R ij ,G ij ,B ij}; The median of the RGB color information of each pixel is mid(R ij ,G ij ,B ij ), the median difference of RGB color information of each pixel is specifically, T ij = mid{|R ij -G ij |,|R ij -B ij |,|G ij -B ij |}, the minimum value of the RGB color information of each pixel is min{R ij ,G ij ,B ij};

[0065] In the embodiment of the present application, the color difference representation value of each pixel point is obtained as follows:

[0066]

[0067] Among them, Rij , G ij , B ij The grayscale values ​​of the three channels R, G, and B corresponding to the pixel points in the i-th row and j-th column are respectively, max is the maximum value in the set, min is the minimum value in the set, and mid is the median value in the set. According to the above formula, the RGB image is calculated point by point to obtain the representation image obtained according to the representation model based on color difference.

[0068] In an embodiment of the present application, by performing color difference characterization processing on each pixel point in the image to be detected, a characterization image is obtained that can more effectively extract targets with color differences in natural scenes, thereby improving the accuracy of extracting suspected sign areas by a fixed threshold segmentation method based on color space conversion.

[0069] Optionally, the binarizing the representation image to obtain a binarized image, and performing denoising on the binarized image to obtain a first suspected target region of the binarized image, comprises:

[0070] Pixels whose color difference representation values ​​are greater than the first preset threshold are regarded as valid pixels, and pixels whose color difference representation values ​​are less than or equal to the first preset threshold are regarded as other pixels;

[0071] A binary image is obtained based on the effective pixel points and other pixel points, and a denoising process is performed on the binary image to obtain a first suspected target area of ​​the binary image.

[0072] Specifically, the binarization in the embodiment of the present application may be to binarize the image using an adaptive first preset threshold, and points with a representation value higher than the first preset threshold are marked as valid points.

[0073] The denoising of the binary image described in the embodiments of the present application specifically involves thinning the low-density areas in the binary image and further discretizing the low-density areas using morphological operations; the denoising of the binary image also includes denoising the thinned image, deleting connected domains that are small in area and that obviously do not have the shape features of the sign, specifically, including connected domains of the sign features whose aspect ratio and duty cycle do not meet the standards, and finally obtaining the first suspected target area of ​​the binary image.

[0074] In the embodiment of the present application, the first suspected target area is effectively identified through binarization processing and denoising processing, which effectively improves the accuracy of identification.

[0075] Optionally, the determining, based on each of the valid pixel points and the first suspected target area, a second suspected target area of ​​the binary image includes:

[0076] Based on the valid pixel point, determine a neighborhood area corresponding to the pixel point;

[0077] Compare the candidate pixels in the neighborhood area with the valid pixels, determine candidate pixels among the candidate pixels, and use the candidate pixels as valid pixels, wherein the candidate pixels in the neighborhood area are pixels having color difference characterization features or RGB color information of the valid pixels;

[0078] All the valid pixels are traversed to determine a second suspected target area of ​​the binary image.

[0079] Specifically, the effective pixel points described in the embodiment of the present application refer to the pixel points whose color difference representation value is higher than the first preset threshold value.

[0080] In an embodiment of the present application, a pixel point whose color difference characterization value is greater than the first preset threshold value is taken as a valid pixel point, and then the valid pixel point in the first suspected target area is used as a seed point. The other non-valid points in the neighborhood area of ​​the point are first compared with the current seed point, and the non-valid points having color difference characterization characteristics and RGB color information characteristics with the valid pixel point are re-marked as valid points. Then, the new valid point is used as the seed point, and regional growth is performed according to the above steps until there are no non-valid points that can continue to grow in the neighborhood of all valid points.

[0081] The color difference characterization features described in the embodiments of the present application and the effective pixel point specifically include the R, G, and B three-channel values ​​of the effective point and the corresponding characterization values ​​based on color differences.

[0082] The candidate pixel points described in the embodiments of the present application are pixel points in the neighborhood that have color difference characterization features of valid pixel points.

[0083] In the embodiment of the present application, through region growing, the first suspected target region can be effectively expanded to obtain the second suspected target region, thereby improving the accuracy of the final recognition result.

[0084] Optionally, determining the connected area corresponding to the road sign based on the color feature confidence and the shape feature confidence of each of the connected areas includes:

[0085] Determine the color feature confidence of each of the connected regions based on the R channel grayscale value, the G channel grayscale value, and the B channel grayscale value of each pixel in the connected region;

[0086] After smoothing the original outer contours of each of the connected regions, a fitting process is performed to obtain the target outer contours of each of the connected regions;

[0087] Determining the shape feature confidence of each of the connected regions based on the similarity between each of the connected regions and the connected region corresponding to the target fitting outer contour;

[0088] Based on the shape feature confidence and the color feature confidence of each of the connected regions, obtaining the target confidence of each of the connected regions;

[0089] Determining the connected area corresponding to the road sign based on the target confidence of each of the connected areas;

[0090] The step of obtaining the target confidence of each of the connected regions is specifically as follows:

[0091]

[0092] Among them, C1 is the color feature confidence, C2 is the shape feature confidence, ω1 and ω2 are the feature allocation weights for the color feature confidence and the shape feature confidence, respectively.

[0093] Specifically, after obtaining the second suspected target area, the color feature information and shape feature information of each connected area in the second suspected target area will be further determined. The color feature information can be the confidence of the connected area in the color feature, and the shape feature information can be the confidence of the connected area in the shape feature.

[0094] Based on the grayscale values ​​of the R channel, the G channel and the B channel of each pixel, the grayscale values ​​of each channel in each pixel are sorted to obtain the grayscale value sorting information of each pixel. The number of categories of grayscale value sorting information is limited. According to different categories of sorting information, the color feature confidence C1 of each connected area is determined.

[0095] More specifically, the confidence of each connected domain in shape features is obtained by first extracting the original outer contour of each connected domain, then smoothing the original outer contour according to the coordinates of the neighborhood points, and finally fitting the processed outer contour coordinate point set. The shape feature confidence C2 of each connected area is determined based on the similarity between the connected domain corresponding to the fitted outer contour and the original connected domain.

[0096] More specifically, based on the shape feature confidence and the color feature confidence of the connected region and the feature weights corresponding to the color feature and the shape feature, the target confidence of each connected region is determined, specifically:

[0097]

[0098] Specifically, ω1 and ω2 are the weights of the color feature confidence and the shape feature confidence, respectively, and C1 and C2 are the color feature confidence and the shape feature confidence, respectively.

[0099] Optionally, determining the color feature confidence of each connected area based on the magnitude of the R channel grayscale value, the G channel grayscale value, and the B channel grayscale value of each pixel point in the connected area includes:

[0100] The grayscale values ​​of the R channel, the G channel and the B channel of each pixel are sorted by size to obtain the grayscale value sorting information of each pixel;

[0101] The color difference representation value corresponding to each of the gray value sorting information is obtained, and the color feature confidence of each of the connected regions is determined based on each of the color difference representation values.

[0102] Specifically, based on each connected domain after region growth, a pixel classification model is established to classify all valid points that constitute the current connected domain. Specifically, the pixel classification model classifies pixels into six categories based on the grayscale values ​​of the three channels R, G, and B of a single pixel, and the priority gradually decreases from top to bottom. The categories are:

[0103]

[0104] Label the category numbers 1-6 one by one from high to low priority for each pixel to obtain the category labeling map of the current connected domain. Calculate the sum of the representation values ​​of all valid points based on color differences in each category, and calculate the total proportion of each category representation value in the current area. The total proportion of the maximum representation value of the connected domain is used as the confidence of the color feature of the connected domain.

[0105] In an embodiment of the present application, the connected areas corresponding to the road signs are determined based on the color feature information and shape feature information of each of the connected areas, thereby solving the problem of misjudgment of the final sign target due to sign deformation, camera acquisition angle, background occlusion and algorithm fitting method.

[0106] Optionally, here, a landscape image containing a circular signboard captured by a vehicle-mounted color camera during driving is taken as an example, and the original image is an RGB image.

[0107] According to formula (1), the original RGB image is characterized based on color difference. The suspected target area extraction includes binarization of the characterized image and denoising of the binarized image. The binarization of the characterized image calculates the mean of the color difference characterization values ​​of the entire image, multiplies the mean by the coefficient K and uses it as the threshold T for binarization, and marks the points above the threshold T as valid points. The denoising of the binarized image first performs an erosion operation on the low-density area in the binarized image and then fills the holes, then deletes the area with a smaller area, and calculates the aspect ratio of the current connected domain and the duty cycle of the corresponding minimum circumscribed rectangle at the same time. The aspect ratio and duty cycle are both in the range of [0.7, 0.9] and are retained.

[0108] For the region growing, for the suspected target region in the binary image, for a single connected domain, all valid points are selected as seed points. Based on the seed point, a neighborhood with a radius of 2 is selected as the growth region. In the original RGB image, the R, G, and B channel values ​​are compared with the channel values ​​of the current seed point. Points with grayscale differences of the corresponding channels less than 5 and absolute differences of the characterization values ​​based on color differences less than 0.3 are marked as new valid points. The new valid points are used as seed points, and the above growth operation is repeated until all valid points are grown and there are no points that meet the conditions in the neighborhood with a radius of 2.

[0109] The confidence level on the color feature is to first take all the valid points in the grown area, classify all the valid points according to formula (3), obtain the category labeling map of the current area, and then calculate the sum of the characterization values ​​of all the valid points under each category based on the color difference according to the category labeling map, and calculate the proportion of each category characterization value to the total characterization value of the current area. The maximum proportion value is taken as the confidence level C1 of the color feature of the connected domain.

[0110] The confidence calculation of each connected domain on the shape feature is performed, the outer contour of each connected domain is extracted, and the coordinates of the middle point are filtered using the coordinates of 7 consecutive points. Then, the Hough circle fitting is performed on the filtered outer contour coordinate point set. According to the center and radius of the fitting circle, the ratio of the number of pixels in the overlapping area of ​​the original connected domain and the fitting circle to the area of ​​the fitting circle is calculated, and the similarity value is used as the confidence C2 of the shape feature.

[0111] The logo area is determined, and for each connected domain after the region growth is completed, the confidence C1 in color features and the confidence C2 in shape of each connected domain are calculated. At the same time, the weight ω1=ω2=0.5 is set, and the confidence C1 and C2 are weighted summed to obtain the total confidence C of the current connected domain. If the total confidence C is greater than 0.6, the current connected domain is considered to be a logo area.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A road sign recognition method, characterized in that: include: Establishing a color difference representation model for each pixel in the image to be detected, so as to obtain a representation image according to the color difference representation model; Binarizing the representation image to obtain a binary image, and performing denoising on the binary image to obtain a first suspected target area of ​​the binary image, wherein the binary image includes M valid pixel points, and the valid pixel points are pixel points whose color difference representation values ​​are higher than a first preset threshold value; Determine a second suspected target region of the binary image based on each of the valid pixel points and the first suspected target region, wherein the second suspected target region includes N connected regions, and M and N are both positive integers; Determining the connected area corresponding to the road sign based on the color feature confidence and shape feature confidence of each of the connected areas; The step of establishing a color difference characterization model for each pixel in the image to be detected to obtain a characterization image according to the color difference characterization model includes: Based on the RGB color information of each pixel point, the maximum value of the RGB color information difference, the median value of the RGB color information and the median value of the RGB color information difference of each pixel point are obtained; Based on the maximum value of the RGB color information difference of the pixel point, the median value of the RGB color information and the median value of the RGB color information difference, a color difference representation model is established to obtain a color difference representation value of each pixel point; Based on each of the color difference representation values, obtaining a representation image; Wherein, determining the second suspected target area of ​​the binary image based on each of the valid pixel points and the first suspected target area includes: Based on the valid pixel point, determine a neighborhood area corresponding to the pixel point; Compare the candidate pixels in the neighborhood area with the valid pixels, determine candidate pixels among the candidate pixels, and use the candidate pixels as valid pixels, wherein the candidate pixels in the neighborhood area are pixels having color difference characterization features or RGB color information features of the valid pixels; All the valid pixels are traversed to determine a second suspected target area of ​​the binary image.

2. The road sign recognition method according to claim 1, characterized in that: The color difference characterization model is specifically: T ij =mid{|R ij -G ij |,|R ij -B ij |,|G ij -B ij |} Among them, R ij , G ij , B ij The grayscale values ​​of the three channels R, G, and B corresponding to the pixel point in the i-th row and j-th column respectively. max is the maximum value in the set, min is the minimum value in the set, and mid is the median value in the set.

3. The road sign recognition method according to claim 1, characterized in that: The binarization process is performed on the representation image to obtain a binarized image, and the denoising process is performed on the binarized image to obtain a first suspected target area of ​​the binarized image, including: Pixels whose color difference representation values ​​are greater than the first preset threshold are regarded as valid pixels, and pixels whose color difference representation values ​​are less than or equal to the first preset threshold are regarded as other pixels; A binary image is obtained based on the effective pixel points and other pixel points, and a denoising process is performed on the binary image to obtain a first suspected target area of ​​the binary image.

4. The road sign recognition method according to claim 1, characterized in that: The method of determining the connected area corresponding to the road sign based on the color feature confidence and the shape feature confidence of each of the connected areas includes: Determine the color feature confidence of each of the connected regions based on the R channel grayscale value, the G channel grayscale value, and the B channel grayscale value of each pixel in the connected region; After smoothing the original outer contours of each of the connected regions, a fitting process is performed to obtain the target outer contours of each of the connected regions; Determining the shape feature confidence of each of the connected regions based on the similarity between the connected regions corresponding to the original outer contours of each of the connected regions and the connected regions corresponding to the target outer contours; Based on the shape feature confidence and the color feature confidence of each of the connected regions, obtaining the target confidence of each of the connected regions; Determining the connected area corresponding to the road sign based on the target confidence of each of the connected areas; The step of obtaining the target confidence of each of the connected regions is specifically as follows: Among them, C1 is the color feature confidence, C2 is the shape feature confidence, ω1 and ω2 are the feature allocation weights for the color feature confidence and the shape feature confidence, respectively.

5. The road sign recognition method according to claim 4, characterized in that: The step of determining the color feature confidence of each connected area based on the R channel grayscale value, the G channel grayscale value, and the B channel grayscale value of each pixel point in the connected area comprises: The grayscale values ​​of the R channel, the G channel and the B channel of each pixel are sorted by size to obtain the grayscale value sorting information of each pixel; The gray value sorting information of each pixel point specifically includes: The color difference representation value corresponding to each of the gray value sorting information is obtained, and the color feature confidence of each of the connected regions is determined based on each of the color difference representation values.

Citation Information

Patent Citations

  • Road sign identification method and road sign identification system

    CN107066933A

  • Traffic sign detection method in natural scene

    CN107301405A