LandPartitioning / Axis.py
Amirhosein
added axis
ef6a23f
import cv2
import numpy as np
from Map import MapIn, CVLineThickness
import random
import config
class AxisFinder:
clicked_points = []
def __init__(self,map:MapIn) -> None:
self.map = map
self.axis_res = []
"""
opencv click callback for customized axis drawing
"""
def click_callback(event, x, y, flags, params):
if event == cv2.EVENT_LBUTTONDOWN:
config.log(f"Clicked Y:{y}, X:{x}")
AxisFinder.clicked_points.append((y,x))
"""
parts parcel number divider based on area of
parts. omits fixed facility areas
"""
def cal_split_parcels(u_map:MapIn,d_map:MapIn,parcels_cnt:int):
u_ff_area = np.sum(u_map.facility_filled_mask)/255
u_block_area = np.sum(u_map.block_mask)/255
u_area = u_block_area - u_ff_area
d_ff_area = np.sum(d_map.facility_filled_mask)/255
d_block_area = np.sum(d_map.block_mask)/255
d_area = d_block_area - d_ff_area
t_area = u_area+d_area
precnt = u_area/t_area
u_parcels = int(parcels_cnt*precnt)
d_parcels = parcels_cnt - u_parcels
return (u_parcels,d_parcels)
"""
sort results by best fitness
"""
def sort_fitness(self,sub_li):
return sub_li.sort(key = lambda x: x[0])
"""
calculates access balance ratio with the provided up&down masks
max value of fitness is 1
"""
def cal_access_split_fitness(self,access_mask:np.ndarray,up_mask:np.ndarray,down_mask:np.ndarray):
img = access_mask
# calculate access ratio
up_sum_mask = up_mask & img
down_sum_mask = down_mask & img
sum_up_access = np.sum(up_sum_mask)/255
sum_down_access = np.sum(down_sum_mask)/255
return 1 - (abs(sum_up_access-sum_down_access)/(np.sum(img)/255))
"""
calculates area balance ratio with provided up&down masks
max value of fitness is 1
"""
def cal_area_split_fitness(self,block_mask:np.ndarray,up_mask:np.ndarray,down_mask:np.ndarray):
imgray = block_mask
up_area = up_mask & imgray
down_area = down_mask & imgray
sum_up_area = np.sum(up_area)/255
sum_down_area = np.sum(down_area)/255
return 1 - (abs(sum_up_area-sum_down_area)/(np.sum(imgray)/255))
"""
calculates fixed facility hit
best answer is 1
"""
def cal_fixed_facilities_fitness(self,facility_mask:np.ndarray,p0:tuple,p1:tuple):
imgray = facility_mask
plain = np.zeros((imgray.shape))
thickness = self.map.roud_thickness + self.map.facility_safe_dist
# input points are (y,x)->(x,y)
plain = cv2.line(plain,(p0[1],p0[0]),(p1[1],p1[0]),255,CVLineThickness.thickness_solver(thickness))
collision = plain.astype(np.uint8) & imgray
max_collision = np.sum(imgray)/255
if max_collision == 0: return 1
collision = np.sum(collision)/255
return 1 - (collision/max_collision)
"""
calculates cut trees with given points
no cut tree = 1
"""
def cal_carbon_fitness(self,tree_mask:np.ndarray,p0:tuple,p1:tuple):
mask = tree_mask
plain = np.zeros((mask.shape))
thickness = self.map.roud_thickness + self.map.tree_safe_dist
# input points are (y,x)->(x,y)
plain = cv2.line(plain,(p0[1],p0[0]),(p1[1],p1[0]),255,CVLineThickness.thickness_solver(thickness))
collision = plain.astype(np.uint8) & mask
# frame must be preprocessed
max_carbon = np.sum(mask)
if max_carbon == 0: return (1,0)
return (1-(np.sum(mask[collision>0])/max_carbon),len(mask[collision>0]))
"""
axis finding fitness function
"""
def fitness_axis(self,solution:tuple,mymap:MapIn,center:tuple):
# solution is y,x of one point
# other point is the center
y_max = mymap.frame_shape[0]-1
x_max = mymap.frame_shape[1]-1
# slope = inf, x=const
if solution[1]==center[1]:
point0 = (0,center[1])
point1 = (y_max,center[1])
# slope = zero, y=const
elif solution[0]==center[0]:
point0 = (center[0],0)
point1 = (center[0],x_max)
# normal line
else:
slope = (solution[0]-center[0])/(solution[1]-center[1])
intercept = center[0] - (slope*center[1])
point0 = (int(intercept),0)
point1 = (int((slope*x_max)+intercept),x_max)
# config.log(point1,point0)
up_mask,down_mask = mymap.line_split_mask_maker(point0,point1)
access_split_fitness = self.cal_access_split_fitness(mymap.access_mask,up_mask,down_mask)
area_split_fitness = self.cal_area_split_fitness(mymap.block_mask,up_mask,down_mask)
fixed_facilities_fitness = self.cal_fixed_facilities_fitness(mymap.fixed_f_mask,point0,point1)
carbon_fitness = self.cal_carbon_fitness(mymap.trees_mask,point0,point1)
# config.log(solution,slope,intercept,point0,point1)
weights = config.A_ACCESS_SPLIT_WEIGHT + config.A_AREA_SPLIT_WEIGHT + config.A_FIXED_FACILITIES_WEIGHT + config.A_CARBON_WEIGHT
score = (config.A_ACCESS_SPLIT_WEIGHT * access_split_fitness + config.A_AREA_SPLIT_WEIGHT * area_split_fitness +
config.A_FIXED_FACILITIES_WEIGHT * fixed_facilities_fitness + config.A_CARBON_WEIGHT * carbon_fitness[0])
# scale score between (0,1]
if fixed_facilities_fitness != 1:
return (-1,point0,point1)
return (score/weights,point0,point1,solution,center,[access_split_fitness,area_split_fitness,fixed_facilities_fitness,carbon_fitness])
"""
iterate through all border pixels and
draw line on start_point to border
finding best line (for the first step of axis finding only)
"""
def iterate_throughall(self,mymap:MapIn, start_point=None):
borders_access = []
if start_point != None:
borders_access = [start_point]
else:
borders_access = mymap.access_mask
borders_access = np.asarray(np.where(borders_access==255))
borders_access = list(zip(borders_access[0], borders_access[1]))
borders_random = mymap.boundry_mask
borders_random = np.asarray(np.where(borders_random==255))
borders_random = list(zip(borders_random[0], borders_random[1]))
borders_random = random.sample(borders_random, int(len(borders_random)*0.4))
res = []
for pixel in borders_access:
for second_point in borders_random:
if np.linalg.norm(np.asarray(pixel)-np.asarray(second_point)) > config.VALID_POINT_DISTANCE:
res.append(self.fitness_axis(pixel,mymap,second_point))
self.sort_fitness(res)
self.axis_res = res[-10:]
return res[-10:]
"""
iterate over old boundries (without division line)
and find best line for New York design
"""
def iterate_old_boundries_new_york(self):
res = []
last_axis_points = self.map.line_points
boundary = self.map.old_boundry_mask
boundary = np.asarray(np.where(boundary==255))
boundary = list(zip(boundary[0], boundary[1]))
# cal perpendicular center (pixelhit line)
for pixel in boundary:
y1, x1 = last_axis_points[0]
y2, x2 = last_axis_points[1]
y3, x3 = pixel
px, py = (x2-x1,y2-y1)
dAB = px*px + py*py
u = ((x3 - x1) * px + (y3 - y1) * py) / dAB
# (y,x)
ppcenter = (int(y1 + u * py),int(x1 + u * px))
if np.linalg.norm(np.asarray(pixel)-np.asarray(ppcenter)) > config.VALID_POINT_DISTANCE:
res.append(self.fitness_axis(pixel,self.map,ppcenter))
self.sort_fitness(res)
self.axis_res = res[-10:]
return res[-10:]