123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100 |
- import numpy as np
- import pandas
- import csv
- from math import sqrt
- from sklearn.cluster import DBSCAN
- import order_bounding_boxes_in_each_block
- def get_average_xy(list_input):
- csv_name = "temporary/list_to_csv_with_corner_points.csv"
- new_list = []
- resultFile = open(csv_name, 'w')
- wr = csv.writer(resultFile, delimiter=";")
- wr.writerow(["element", "xmin","ymin","xmax","ymax", "ausrichtung","point_xmi_ymi","point_xma_ymi","point_xmi_yma","point_xma_yma"])
- for element in list_input:
- xavg_elem = 0
- yavg_elem = 0
- ymin = 100000000
- ymax = 0
- xmin = 100000000
- xmax = 0
- for blub in element: #get the smallest and largest x and y value for whole block
- xavg_elem += (float(blub[0]) + float(blub[2]))/2
- yavg_elem += (float(blub[1]) + float(blub[3]))/2
- if float(blub[1]) < ymin:
- ymin = float(blub[1])
- #print("y_min:",y_min)
- if float(blub[0]) < xmin:
- xmin = float(blub[0])
- if float(blub[3]) > ymax:
- ymax = float(blub[3])
- if float(blub[2]) > xmax:
- xmax = float(blub[2])
- if xmax-xmin > ymax-ymin:
- ausrichtung = 0 #horizontal
- else:
- ausrichtung = 1 #vertikal
- xavg_elem = xavg_elem/len(element)
- #print(xavg_elem)
- yavg_elem = yavg_elem/len(element)
- #element.extend([xavg_elem, yavg_elem])
- #print(element)
- #new_list.append(element)
- ##### GET CORNER POINTS
- point_xmi_ymi = [xmin,ymin]
- point_xma_ymi = [xmax,ymin]
- point_xmi_yma = [xmin,ymax]
- point_xma_yma = [xmax,ymax]
- wr.writerow([element,xmin,ymin,xmax,ymax, ausrichtung,point_xmi_ymi,point_xma_ymi,point_xmi_yma,point_xma_yma])
- resultFile.close()
- #print(new_list)
- return csv_name
- def dist(rectangle1, rectangle2):
- #get minimal distance between two rectangles
- distance = 100000000
- #print(rectangle1)
- for point1 in rectangle1:
- point1 = eval(point1)
- #print(point1)
- for point2 in rectangle2:
- #print(point2)
- point2 = eval(point2)
- dist = sqrt((float(point2[0]) - float(point1[0]))**2 + (float(point2[1]) - float(point1[1]))**2)
- if dist < distance:
- distance = dist
- return distance
- def clustering(distance_matrix):
- db = DBSCAN(eps=5, min_samples=1, metric="precomputed").fit(dm) ##3.93 until now, bei 5 shon mehr erkannt, 7 noch mehr erkannt aber auch schon zu viel
- labels = db.labels_
- # Number of clusters in labels
- n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
- print('Estimated number of clusters: %d' % n_clusters_)
- data_df = pandas.read_csv("/home/bscheibel/PycharmProjects/dxf_reader/temporary/list_to_csv_with_corner_points.csv",
- sep=";")
- data_df["cluster"] = labels
- data_df.groupby('cluster')['element'].apply(' '.join).reset_index().to_csv("values_clusteredfrom_precomputed_dbscan.csv",sep=";")
- file = "/home/bscheibel/PycharmProjects/dxf_reader/drawings/5152166_Rev04.html"
- result = order_bounding_boxes_in_each_block.get_bound_box(file)
- #print(result)
- get_average_xy(result)
- #rectangle1 = [[0,0],[2,0],[0,2],[2,2]]
- #rectangle2 = [[3,3],[4,3],[3,4],[4,4]]
- #print(compute_distance(rectangle1,rectangle2))
- data = pandas.read_csv("/home/bscheibel/PycharmProjects/dxf_reader/temporary/list_to_csv_with_corner_points.csv", sep=";")
- data = data[["point_xmi_ymi","point_xma_ymi","point_xmi_yma","point_xma_yma"]].replace("'","")
- #print(data)
- data.to_csv("blub.csv", sep=";", index=False, header=None)
- result = []
- with open('blub.csv') as csvfile:
- readCSV = csv.reader(csvfile, delimiter=';')
- result = list(readCSV)
- dm = np.asarray([[dist(p1, p2) for p2 in result] for p1 in result])
- #print(dm)
- clustering(dm)
|