(a) (b) (c)
Figure 13. Partial view of line detection
This method is compared with the original k-means clustering results, as shown in Figure 14. Figure 14a shows the clustering results of the original k-means algorithm in the Hough space for the fitted lines at thresholds of 25, 30, and 35. Use circles and oblique crosses to represent the centroids of k-means clustering, and data points in different clusters are represented in different colors. In Figure 14, when is 25, due to the presence of many interfering line parameter points in the Hough space, the clustering results are severely affected by interference, and the interfering points are mistakenly clustered into one category; When is 30, due to the increase in threshold, some of the interference line parameter points in the Hough space have been deleted, but there is still some interference, resulting in inaccurate clustering results; When is 35, the interfering line parameter points are further deleted, and the line parameter points within each cluster happen to be accurate lines, so the clustering results are good. Using the adaptive threshold proposed in this article combined with an improved clustering centroid calculation method, In Figure 14b, the method of determining the optimal adaptive threshold based on statistical voting numbers adaptively sets according to the distribution of voting numbers, removes the interfering line parameter points, and then recalculates the clustering centroid according to the weight proportion of voting numbers in the clustering results. When the Hough transform threshold is 25, 30, and 35, clustering analysis can be completed correctly. Overall, the clustering robustness and accuracy of the method proposed in this article are superior to the original k-means clustering. The calculation flowchart is shown in Figure 15.