Search

CN-121982544-A - Adaptive K-means road segmentation method based on priori guidance

CN121982544ACN 121982544 ACN121982544 ACN 121982544ACN-121982544-A

Abstract

The invention relates to a priori guidance-based self-adaptive K-means road segmentation method, belongs to the technical field of road segmentation, and solves the technical problems that the prior road segmentation technology adopts fixed parameter priori and is difficult to adapt to dynamic change of battlefield environment. The self-adaptive K-means road segmentation method based on priori guidance comprises the following steps of S1, obtaining an original image of a road and preprocessing, S2, defining a trapezoid road priori region based on brightness, defining an inverted concave background priori region based on saturation, setting an initial clustering center, S3, adopting a multi-level sub-sampling contour coefficient method to self-adaptively determine an optimal K value, selecting the optimal K value as the optimal clustering number and carrying out K-means clustering segmentation, S4, mapping a clustering result to a semantic category through an priori center matching mechanism, and obtaining a final segmented image through morphological processing. Compared with the prior art, the method has the advantages of parameter self-adaption, high segmentation precision, strong adaptability to complex road environments and the like.

Inventors

  • ZHAO LICHEN
  • WU SHUAIYI
  • LI ZHENGFEI
  • LI XINGMIN
  • YAN YONG

Assignees

  • 中北大学

Dates

Publication Date
20260505
Application Date
20260130

Claims (9)

  1. 1. The adaptive K-means road segmentation method based on priori guidance is characterized by comprising the following steps of: S1, acquiring an original image of a road, and preprocessing the original image of the road through an improved MSRCR method, HSV color space conversion and contrast limited self-adaptive histogram equalization; s2, defining a trapezoid road prior area based on brightness, and defining an inverted concave background prior area based on saturation; s3, adopting a multi-level sub-sampling contour coefficient method to adaptively determine an optimal K value, selecting the optimal K value as the optimal clustering quantity, and carrying out K-means clustering segmentation; And S4, mapping the clustering result to a semantic category through a priori center matching mechanism, and obtaining a final segmentation image through morphological processing.
  2. 2. The adaptive K-means road segmentation method based on prior guidance of claim 1, wherein the method for preprocessing the original image of the road by the improved MSRCR method in the step S1 is as follows: By passing through Calculation of , For the reflection component estimate of the pixel (x, y) on channel p after multi-scale Retinex enhancement, The image pixel value of the p-th channel in the original image of the road, The result of the p-th channel in the original image of the road after the q-th scale Gaussian filtering, ; By passing through Calculation of , For the color recovery factor at pixel (x, y) on channel p, , In order to be an intensity image, For the gain parameter, 125, Taking 0.1 for scaling parameters; By passing through Calculation of , Is a color restored multi-scale retinal enhancement value at pixel (x, y) on channel p; Will be According to Normalization processing to obtain , Is a multi-scale retinal enhancement value normalized at pixel (x, y) on channel p and color restored.
  3. 3. The method for adaptively partitioning K-means roads based on prior guidance as set forth in claim 1, wherein said preprocessing by HSV color space conversion in step S1 is to separate color information from luminance information.
  4. 4. The adaptive K-means road segmentation method based on prior guidance of claim 1, wherein the method for preprocessing the original image of the road through contrast-limited adaptive histogram equalization in the step S1 is as follows: Dividing an image into 8 x 8 regions, and performing local histogram equalization with a contrast limit of 0.02 on each region; By passing through Each of the regions is cut out and, , In order to tailor the clipping parameters, For the total number of pixels per region, L is the number of gray levels, 256 is taken, An upper limit value for the number of pixels allowed for any gray level in the region histogram, For the histogram of each region, Intermediate histograms before pixel redistribution for applying clipping threshold; According to Calculating the number of excess pixels after each region is cut ; According to The excess number of pixels is assigned to all gray levels, The reassigned histogram is uniform for the clipped pixels.
  5. 5. The adaptive K-means road segmentation method based on prior guidance according to claim 1, wherein in the step S2, when the brightness is high, the road height is 0.6, the background height is 0.4/0.25, when the brightness is low, the road height is 0.4, the background height is 0.5/0.3, when the saturation is high, the road height is 0.4, the background height is 0.4/0.25, and when the saturation is low, the road height is 0.4, and the background height is 0.5/0.3.
  6. 6. The adaptive K-means road segmentation method based on prior guidance of claim 4, wherein the method for adaptively determining the optimal K value by using the multi-level sub-sampling contour coefficient method in the step S3 is as follows: constructing an evaluation set E from the original data set D according to The amount of the sample is determined and, S is the total number of samples; performing candidate cluster number K value Sub-independent sub-sampling, extracting sub-sampling set from evaluation set E According to Determining sub-sample size ; According to The subset average profile coefficients are calculated and, , For the contour coefficients of the sample point i, , , For the average euclidean distance of sample i from sample j, For the average distance of a sample point from other sample points of the class to which it belongs, For the average distance of a sample point from sample points in other nearest clusters, Is a cluster Is used for the number of sample points of (a), To remove The number of sample points of other clusters outside; For the sub-sampled sample size, In the v-th sampling, the sub-sampling set The contour coefficient calculated for the ith sample, The average profile coefficient of the subset corresponding to the v sub-sampling is obtained; Execution of After sub-independent sub-sampling, according to And calculating the approximate contour coefficient of the candidate cluster number K value to obtain the optimal K value of 4.
  7. 7. The adaptive K-means road segmentation method based on prior guidance of claim 6, wherein the method for selecting the optimal K value as the optimal cluster number and performing K-means cluster segmentation in the step S3 is as follows: S11, confirming that the optimal clustering quantity is 4, and setting the average value of the prior area of the trapezoid road and the prior area of the inverted concave background as an initial clustering center; S12, calculating Euclidean distances from all sample points to an initial clustering center, dividing the Euclidean distances into clusters which are located at the closest clustering center, and calculating the average value of all sample points in the clusters as a new clustering center; s13, repeating the operation S12 until the maximum iteration times or the cluster center is not changed; S14, outputting a moving track graph of the clustering center and a final segmentation result graph.
  8. 8. The method for adaptively partitioning a K-means road based on prior guidance as set forth in claim 1, wherein said step S4 is performed according to the following 、 A mapping of cluster centers to semantic categories is established, For the result of the road class mapping, For the center of the ith cluster, Is the prior center of the trapezoid road, As a result of the mapping of the background category, Is the prior center of the 'inverted concave' background.
  9. 9. The method for adaptively segmenting the K-means road based on prior guidance according to claim 1, wherein the morphological processing in the step S4 comprises binarizing the image, performing an open operation, removing small areas and removing protrusions.

Description

Adaptive K-means road segmentation method based on priori guidance Technical Field The invention belongs to the technical field of road segmentation, and particularly relates to a self-adaptive K-means road segmentation method based on priori guidance. Background The intelligent unmanned combat vehicle (UGCV) is used as one of the core equipment of future intelligent warfare, and the environment perception and the passable road segmentation capability are the basic key links for realizing autonomous navigation and tactical maneuver. Different from the urban structured road environment, the intelligent unmanned combat vehicle combat environment has the remarkable characteristics of extremely unstructured, high dynamic and strong anti-interference, and is particularly characterized by severe illumination change, complex and diversified landforms and severe requirements on real-time, and the factors together form a serious technical challenge. In the development process of the road segmentation technology, three technical routes, namely a traditional image processing method, a method based on supervised learning and an unsupervised/light supervised learning method, are mainly formed. Along with the development of deep learning technology, a road segmentation algorithm which does not depend on a large amount of annotation data becomes an important research direction. In the unsupervised learning method, the K-means clustering algorithm is paid attention to widely because of its feature of no need of training. However, the method still has two technical bottlenecks that the clustering number K needs to be preset and cannot be adjusted in a self-adaptive mode, and low-level visual features lack semantic association. At present, studies are made to remedy semantic deficiency by introducing priori knowledge. However, most of these methods use fixed parameter priors, which are difficult to adapt to dynamic changes in the battlefield environment. Disclosure of Invention In order to overcome the defects of the prior art and solve the technical problems that fixed parameter priori is adopted in the prior road segmentation technology, dynamic change of battlefield environment is difficult to adapt to and the like, the invention provides a priori guidance-based self-adaptive K-means road segmentation method. The invention is realized by the following technical scheme. The invention provides a priori guidance-based self-adaptive K-means road segmentation method, which comprises the following steps: S1, acquiring an original image of a road, and preprocessing the original image of the road through an improved MSRCR method, HSV color space conversion and contrast limited self-adaptive histogram equalization; s2, defining a trapezoid road prior area based on brightness, and defining an inverted concave background prior area based on saturation; s3, adopting a multi-level sub-sampling contour coefficient method to adaptively determine an optimal K value, selecting the optimal K value as the optimal clustering quantity, and carrying out K-means clustering segmentation; And S4, mapping the clustering result to a semantic category through a priori center matching mechanism, and obtaining a final segmentation image through morphological processing. Further, in the step S1, the method for preprocessing the original image of the road by the improved MSRCR method includes: By passing through Calculation of,For the reflection component estimate of the pixel (x, y) on channel p after multi-scale Retinex enhancement,The image pixel value of the p-th channel in the original image of the road,The result of the p-th channel in the original image of the road after the q-th scale Gaussian filtering,; By passing throughCalculation of,For the color recovery factor at pixel (x, y) on channel p,,In order to be an intensity image,For the gain parameter, 125,Taking 0.1 for scaling parameters; By passing through Calculation of,Is a color restored multi-scale retinal enhancement value at pixel (x, y) on channel p; Will be According toNormalization processing to obtain,Is a multi-scale retinal enhancement value normalized at pixel (x, y) on channel p and color restored. Further, the preprocessing of HSV color space conversion in step S1 is to separate color information from luminance information. Further, in the step S1, the method for preprocessing the original image of the road through the contrast limited adaptive histogram equalization includes: Dividing an image into 8 x 8 regions, and performing local histogram equalization with a contrast limit of 0.02 on each region; By passing through Each of the regions is cut out and,,In order to tailor the clipping parameters,For the total number of pixels per region, L is the number of gray levels, 256 is taken,An upper limit value for the number of pixels allowed for any gray level in the region histogram,For the histogram of each region,Intermediate histograms before pixel redistribution for applying