Amirhosein
First try
eb9da7f
import cv2
import numpy as np
import pandas as pd
import random
from functools import singledispatchmethod
import math
import os
from sympy import false
import config
class MapIn:
# thickness configs
tree_safe_dist = config.TREE_SAFE_DIST
facility_safe_dist = config.FACILITY_SAFE_DIST
# parcel min area
parcel_minimum_area = config.AXIS_MIN_AREA
# ratio of access to boundry to stop halving
access_ratio = config.ACCESS_RATIO
"""
map class to handle raw inputs
RGB (round access, green factor, boundary)
Background is white by deafult
Black shows fixed facilities
"""
@singledispatchmethod
def __init__(self) -> None:
assert False, 'bad input'
@__init__.register(str)
def _first__(self,src:str,src_block:str,src_ff:str,parcel_cnt:int,arch_choice:config.ArchStyles,map_id:int) -> None:
self.roud_thickness = config.ROAD_SIZE_MAX
self.map_id = map_id
self.frame = cv2.imread(src)
self.frame_shape = self.frame.shape
self.arch_choice = arch_choice
self.parcel_cnt = parcel_cnt
self.centers = (int(self.frame_shape[0]/2),int(self.frame_shape[1]/2))
self.trees_mask, self.fixed_f_mask,self.access_mask, self.boundry_mask = self.create_masks()
self.trees_binary_mask = cv2.threshold(self.trees_mask, 127, 255, cv2.THRESH_BINARY)[1]
# read block mask
self.block_mask = cv2.imread(src_block)
self.block_mask = cv2.cvtColor(self.block_mask,cv2.COLOR_BGR2GRAY)
# read facility filled mask
self.facility_filled_mask = cv2.imread(src_ff)
self.facility_filled_mask = cv2.cvtColor(self.facility_filled_mask,cv2.COLOR_BGR2GRAY)
self.print_report()
# Axis maps initialization
@__init__.register(np.ndarray)
def _second__(self,split_mask:np.ndarray,parent_map,line_mask:np.ndarray,map_id:int,dir:int,line_p) -> None:
self.line_p = line_p
# self.axis_center = parent_map.axis_center
self.split_mask = split_mask
self.dir = dir # 0 up 1 down
self.roud_thickness = config.ROAD_SIZE_MAX - config.ROAD_STEP*int(math.log(map_id+1,2))
if self.roud_thickness <= config.ROAD_STEP: self.roud_thickness=config.ROAD_SIZE_MIN
config.log(f"roud thickness:{self.roud_thickness} map_id:{map_id}")
self.map_id = map_id
# split_mask_3d = np.zeros((self.frame_shape))
# self.frame = parent_map.frame & split_mask
split_3d_mask = np.zeros(parent_map.frame_shape, dtype=np.uint8)
split_3d_mask[:,:,:] = split_mask[:,:,np.newaxis]
self.frame = parent_map.frame & split_3d_mask
self.frame_shape = self.frame.shape
self.arch_choice = parent_map.arch_choice
self.centers = (int(self.frame_shape[0]/2),int(self.frame_shape[1]/2))
self.trees_mask = parent_map.trees_mask & split_mask
self.trees_binary_mask = parent_map.trees_binary_mask & split_mask
self.fixed_f_mask = parent_map.fixed_f_mask & split_mask
self.boundry_mask = parent_map.boundry_mask & split_mask
self.old_boundry_mask = parent_map.boundry_mask & split_mask
self.block_mask = parent_map.block_mask & split_mask
self.facility_filled_mask = parent_map.facility_filled_mask & split_mask
# add new line to access
new_access_line = self.block_mask & line_mask
self.access_mask = parent_map.access_mask & split_mask
self.access_mask = self.access_mask | new_access_line
# add new line to boundries
self.boundry_mask = self.boundry_mask | new_access_line
self.new_access_line = new_access_line
if map_id > 2:
self.parent_access_line = parent_map.new_access_line
self.parent_line_p = parent_map.line_p
else:
self.parent_access_line = self.new_access_line
self.parent_line_p = self.line_p
self.save_map()
self.print_report()
# Parcels initilization
@__init__.register(int)
def _third__(self,parcel_id:int,split_mask:np.ndarray,parent_map,parcel_area,lines_points_tup,parcel_type) -> None:
self.dir = parent_map.dir
self.parent_line_p = parent_map.parent_line_p
self.line_p = parent_map.line_p
self.parent_access_line = parent_map.parent_access_line & split_mask
self.access_line = parent_map.new_access_line & split_mask
self.parcel_id = parcel_id
self.curr_size = parcel_area
self.bounding_lines = lines_points_tup
self.parcel_type = parcel_type
self.map_id = parent_map.map_id
split_3d_mask = np.zeros(parent_map.frame_shape, dtype=np.uint8)
split_3d_mask[:,:,:] = split_mask[:,:,np.newaxis]
self.frame = parent_map.frame & split_3d_mask
self.frame_shape = self.frame.shape
self.arch_choice = parent_map.arch_choice
self.trees_mask = parent_map.trees_mask & split_mask
self.trees_binary_mask = parent_map.trees_binary_mask & split_mask
self.fixed_f_mask = parent_map.fixed_f_mask & split_mask
self.boundry_mask = parent_map.boundry_mask & split_mask
self.block_mask = parent_map.block_mask & split_mask
self.facility_filled_mask = parent_map.facility_filled_mask & split_mask
# make center of new parcel
block_mask = self.block_mask.astype(np.uint8)
contours, _ = cv2.findContours(block_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
cnts = sorted(contours, key=cv2.contourArea, reverse=True)
M = cv2.moments(cnts[0])
cX = int(M["m10"] / M["m00"])
cY = int(M["m01"] / M["m00"])
self.parcel_center = (cY,cX)
# self.save_map()
self.print_report()
def print_report(self):
config.log(f"Map {self.map_id} Area : {np.sum(self.block_mask)/255}")
config.log(f"Map {self.map_id} Tree Area : {np.sum(self.trees_binary_mask)/255}")
config.log(f"Map {self.map_id} Fixed-Facility Area : {np.sum(self.facility_filled_mask)/255}")
config.log(f"Map {self.map_id} Sparse Area : {np.sum(self.block_mask & np.bitwise_not(self.facility_filled_mask) & np.bitwise_not(self.trees_binary_mask))/255}")
def set_map_axis_center(self,point):
self.axis_center = point
def set_line_point(self,point:tuple):
self.line_points = point
def save_map(self) -> None:
if config.WRITE_UNNECESSARY:
cv2.imwrite(f'outputs/map{self.map_id}.bmp',self.frame)
cv2.imwrite(f'outputs/access_mask{self.map_id}.bmp',self.access_mask)
cv2.imwrite(f'outputs/boundry_mask{self.map_id}.bmp',self.boundry_mask)
def create_masks(self) -> tuple:
res = []
img_re = self.frame.reshape(-1,3)
df = pd.DataFrame(img_re,columns=['b','g','r'])
df['r'].astype(np.uint8)
df['g'].astype(np.uint8)
df['b'].astype(np.uint8)
indx_trees = df.apply(lambda x: x.b==0 and 0<x.g<=255 and x.r==0, axis=1)
df_trees = df.copy()
df_trees[np.logical_not(indx_trees)] = [0,0,0]
# df_trees[indx_trees] = [255,255,255]
out = df_trees.values.reshape(self.frame_shape)
out = out.astype(np.uint8)
out[:,:,0] = 0
out[:,:,2] = 0
out = cv2.threshold(out, 127, 255, cv2.THRESH_BINARY)[1][:,:,1]
res.append(out)
cv2.imwrite('outputs/tree_mask.bmp',out)
indx_fixed_fac = df.apply(lambda x: x.b==0 and x.g==0 and x.r==0, axis=1)
df_ff = df.copy()
df_ff[np.logical_not(indx_fixed_fac)] = [0,0,0]
df_ff[indx_fixed_fac] = [255,255,255]
out = df_ff.values.reshape(self.frame_shape)
out = out.astype(np.uint8)
out = cv2.cvtColor(out,cv2.COLOR_BGR2GRAY)
res.append(out)
cv2.imwrite('outputs/facility_mask.bmp',out)
indx_access = df.apply(lambda x: x.g==0 and x.r==255, axis=1)
df_ac = df.copy()
df_ac[np.logical_not(indx_access)] = [0,0,0]
df_ac[indx_access] = [255,255,255]
out = df_ac.values.reshape(self.frame_shape)
out = out.astype(np.uint8)
out = cv2.cvtColor(out,cv2.COLOR_BGR2GRAY)
res.append(out)
cv2.imwrite('outputs/access_mask.bmp',out)
indx_boundry = df.apply(lambda x: x.b==255 and x.g==0, axis=1)
df_b = df.copy()
df_b[np.logical_not(indx_boundry)] = [0,0,0]
df_b[indx_boundry] = [255,255,255]
out = df_b.values.reshape(self.frame_shape)
out = out.astype(np.uint8)
out = cv2.cvtColor(out,cv2.COLOR_BGR2GRAY)
res.append(out)
cv2.imwrite('outputs/boundary_mask.bmp',out)
return tuple(res)
def correct_input(self) -> None:
img_re = self.frame.reshape(-1,3)
df = pd.DataFrame(img_re,columns=['b','g','r'])
df['r'].astype(np.uint8)
df['g'].astype(np.uint8)
df['b'].astype(np.uint8)
# ----set tree values to random
random.seed(13)
df = df.apply(lambda x: [0,random.randint(1,255),0] if x['b']==0 and x['g']==255 and x['r']==0 else x,axis=1)
# ----convert white to black
# indx = df.apply(lambda x: x['b']==255 and x['g']==255 and x['r']==255,axis=1)
# df[indx] = [0,0,0]
# ----set rgb(0,255,255) to rgb(255,0,255)
# indx = df.apply(lambda x: x['b']==255 and x['g']==255 and x['r']==0,axis=1)
# df[indx] = [255,0,255]
# ----save output
out = df.values.reshape(self.frame_shape)
out = out.astype(np.uint8)
cv2.imwrite('outputs/kan_pre.bmp',out)
self.frame = out
# ----print
# cv2.imshow("kan_pre_out", out)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
"""
returns above the line mask and below the line mask
"""
def line_split_mask_maker(self,p0:tuple,p1:tuple):
# points are (y,x) oriented
img_pixels = self.frame_shape[0]*self.frame_shape[1]
img_x = self.frame_shape[1]
# numpy image is (y*x*3)
# !! unit16 may not be enough
y_index = (np.arange(img_pixels).reshape(self.frame_shape[:2])/img_x).astype(np.uint32)
x_index = np.arange(img_pixels).reshape(self.frame_shape[:2])%img_x
if p1[1] == p0[1]:
up_down_line = x_index - p0[1]
else:
slope = (p1[0]-p0[0])/(p1[1]-p0[1])
intercept = p0[0] - (slope*p0[1])
up_down_line = x_index*slope + intercept - y_index
# down part (becareful about center)
down_mask = np.where(up_down_line>=0,255,0).reshape(self.frame_shape[:2])
up_mask = np.where(up_down_line>=0,0,255).reshape(self.frame_shape[:2])
return (up_mask,down_mask)
"""
returns only line mask on main image
"""
def line_mask_maker(self,p0:tuple,p1:tuple):
plain = np.zeros((self.block_mask.shape))
plain = cv2.line(plain,(p0[1],p0[0]),(p1[1],p1[0]),255,2)
return plain.astype(np.uint8)
"""
check whether the half map has a feasible condition
or supports the finishing condtion.
"""
def isfeasible(self):
# check if the part has more than 60% access
access = np.sum(self.access_mask)/255
boundry = np.sum(self.boundry_mask)/255
access_ratio = access/boundry
access_cond = access_ratio<self.access_ratio
# area of part not smaller than standard
block_size = np.sum(self.block_mask)/255
size_cond = block_size>self.parcel_minimum_area
config.log(f'block size:{block_size} access_ratio:{access_ratio} map_id:{self.map_id}')
self.curr_access = access_ratio
self.curr_size = block_size
return access_cond and size_cond
class CVLineThickness:
"""
method selects cv2line arg
depending on the pixel width
"""
@staticmethod
def thickness_solver(desired_thickness):
if desired_thickness == 1:
return 1
if desired_thickness == 2:
# assert false, f"change road size or road step to odd number: {desired_thickness}"
return 2
if desired_thickness == 3:
return 2
if desired_thickness % 2 == 0:
# assert false, f"change road size or road step to odd number: {desired_thickness}"
return desired_thickness - 1
return desired_thickness - 2
class MapOut:
def __init__(self,src:str,lines_axis:list) -> None:
self.img = cv2.imread(src)
self.img_axised = self.img.copy()
self.img_partitioned = None
self.img_built = None
self.img_last = None
self.axis_lines = lines_axis
self.partitioning_lines = None
self.parcels_dic = {}
self.building_masks = None
# export details
self.total_carbon = 0
self.total_trees = 0
self.total_carbon_loss = 0
self.total_cut_tree = 0
self.total_axis_length = 0
self.total_axis_per_block_pr = 0
self.total_num_parcels = 0
self.total_num_parcels_types = {p_type:0 for p_type in config.ParcelType._member_names_}
self.total_sum_ff = 0
def reset_map_for_partitioning(self):
self.total_num_parcels = 0
self.total_num_parcels_types = {p_type:0 for p_type in config.ParcelType._member_names_}
self.total_sum_ff = 0
self.partitioning_lines = None
if self.img_partitioned is not None:
self.img_partitioned = None
self.img_last = self.img_axised.copy()
def reset_map_for_location_finding(self):
self.total_sum_ff = 0
self.building_masks = None
if self.img_built is not None:
self.img_built = None
self.img_last = self.img_partitioned.copy()
def add_partition_report(self,report):
self.total_num_parcels += report['cnt']
report.pop('cnt')
for p_type in report.keys():
self.total_num_parcels_types[p_type] += report[p_type]
def report(self):
self.block_mask = cv2.imread(config.MAIN_MAP_FILLED_BLOCK_MASK)
self.tree_mask = cv2.imread('outputs/tree_mask.bmp')
self.binary_tree_mask = cv2.threshold(self.tree_mask, 127, 255, cv2.THRESH_BINARY)[1]
self.facility_filled_mask = cv2.imread(config.MAIN_MAP_FILLED_F_F_MASK)
# total trees carbon
self.total_carbon = np.sum(self.tree_mask)/3
self.total_trees = np.sum(self.binary_tree_mask)/(255*3)
# calculate tree cut and carbon
self.img_last = self.img_last & self.block_mask
self.img_last = self.img_last.astype(np.uint8)
self.img_mask = cv2.threshold(self.img_last, 127, 255, cv2.THRESH_BINARY)[1]
# omit roads mask and building mask from it
self.roads_mask = cv2.imread('outputs/roads_mask.bmp')
collision3dmask = self.roads_mask
if os.path.exists('outputs/buildings_mask.bmp'):
collision3dmask = collision3dmask | cv2.imread('outputs/buildings_mask.bmp')
if os.path.exists('outputs/partitioning_mask.bmp'):
collision3dmask = collision3dmask | cv2.imread('outputs/partitioning_mask.bmp')
cv2.imwrite('outputs/constructed_mask.bmp', collision3dmask)
self.total_carbon_loss = np.sum(collision3dmask & self.tree_mask)/3
self.total_cut_tree = np.sum(collision3dmask & self.binary_tree_mask)/(255*3)
config.log(f"Total Trees:{self.total_trees} Total Carbon Values:{self.total_carbon}")
config.log(f"Total Cut Trees:{self.total_cut_tree} Total Carbon Loss:{self.total_carbon_loss}")
config.log(f"Total Cut Precentage:{self.total_cut_tree/self.total_trees}")
# calculate axis reports
img_re = self.img_last.reshape(-1,3)
df = pd.DataFrame(img_re,columns=['b','g','r'])
df['r'].astype(np.uint8)
df['g'].astype(np.uint8)
df['b'].astype(np.uint8)
indx_axis = df.apply(lambda x:x.g == 0 and 0<x.r<=255 and x.b == 0, axis=1)
df_axis = df.copy()
df_axis[np.logical_not(indx_axis)] = [0,0,0]
out = df_axis.values.reshape(self.img_last.shape)
out = out.astype(np.uint8)
out = cv2.cvtColor(out,cv2.COLOR_BGR2GRAY)
out = cv2.threshold(out, 1, 255, cv2.THRESH_BINARY)[1]
self.total_axis_length = np.sum(out)/255
self.total_axis_per_block_pr = self.total_axis_length / (np.sum(self.block_mask)/(255*3))
sparse_area = np.sum(self.block_mask & np.bitwise_not(self.facility_filled_mask) & np.bitwise_not(self.binary_tree_mask))/(255*3)
self.total_axis_per_sparse_pr = self.total_axis_length / sparse_area
config.log(f"Total Axis Area:{self.total_axis_length}")
config.log(f"Total Block Area:{np.sum(self.block_mask)/(255*3)}")
config.log(f"Total Sparse Area:{sparse_area}")
config.log(f"Total Axis Per Block Precentage:{self.total_axis_per_block_pr}")
config.log(f"Total Axis Per Sparse Precentage:{self.total_axis_per_sparse_pr}")
# Parcels number plus types
config.log(f"Total Parcels:{self.total_num_parcels}")
config.log(f"Total Parcel types:{self.total_num_parcels_types}")
config.log(f"Total Parcels With FF:{self.total_sum_ff}")
def draw_axis(self):
for line in self.axis_lines:
p0=line[0][0]
p1=line[0][1]
thickness=config.ROAD_SIZE_MAX - config.ROAD_STEP*int(math.log(line[1]+1,2))
if thickness <= config.ROAD_SIZE_MIN:
thickness=config.ROAD_SIZE_MIN
# draw line of axis
self.img_axised = cv2.line(self.img_axised,(p0[1],p0[0]),(p1[1],p1[0]),(0,0,127),CVLineThickness.thickness_solver(thickness+2))
# rotate points
self.img_axised = cv2.line(self.img_axised,(p0[1],p0[0]),(p1[1],p1[0]),(0,0,255),CVLineThickness.thickness_solver(thickness))
cv2.imwrite('outputs/final_axis.bmp',self.img_axised)
# save road lines mask
img_re = self.img_axised.reshape(-1,3).copy()
df = pd.DataFrame(img_re,columns=['b','g','r'])
df['r'].astype(np.uint8)
df['g'].astype(np.uint8)
df['b'].astype(np.uint8)
indx_axis = df.apply(lambda x:x.g == 0 and 0<x.r<=255 and x.b == 0, axis=1)
df[np.logical_not(indx_axis)] = [0,0,0]
out = df.values.reshape(self.img_axised.shape)
out = out.astype(np.uint8)
out = cv2.cvtColor(out,cv2.COLOR_BGR2GRAY)
out = cv2.threshold(out, 1, 255, cv2.THRESH_BINARY)[1]
self.img_last = self.img_axised.copy()
cv2.imwrite('outputs/roads_mask.bmp', out)
def draw_partitions(self,iteration:int,map:MapIn,lines_parcels:list):
map_id = map.map_id
if lines_parcels is not None:
self.parcels_dic[map_id] = lines_parcels
lines_mask = np.zeros(self.img_last.shape, dtype=np.uint8)
for line in lines_parcels:
p0=line[0]
p1=line[1]
# rotate points
lines_mask = cv2.line(lines_mask,(p0[1],p0[0]),(p1[1],p1[0]),(120,120,120),1)
split_3d_mask = np.zeros(self.img_last.shape, dtype=np.uint8)
split_3d_mask[:,:,:] = map.block_mask[:,:,np.newaxis]
lines_mask = lines_mask & split_3d_mask
self.img_partitioned = self.img_last.astype(np.uint8) & np.bitwise_not(lines_mask)
# write partitioning mask
lines_mask = np.where(lines_mask>0,(255,255,255), (0,0,0))
if self.partitioning_lines is not None:
self.partitioning_lines |= lines_mask
else:
self.partitioning_lines = lines_mask
if config.WRITE_UNNECESSARY:
cv2.imwrite(f'outputs/final_map_{iteration}_{map_id}.bmp',self.img_partitioned)
self.img_last = self.img_partitioned.copy()
def draw_partitioning_results(self):
cv2.imwrite(f'outputs/partitioning_mask.bmp',self.partitioning_lines)
cv2.imwrite(f'outputs/final_map_partitioning.bmp',self.img_partitioned)
def draw_building(self,building_mask,iteration,map_id,parcel_id,has_building,parcel_type,block_mask):
color3d_mask = np.zeros(self.img_last.shape, dtype=np.uint8)
color3d_mask[:,:,:] = block_mask[:,:,np.newaxis]
if has_building:
self.total_sum_ff += 1
color3d_mask = np.where(color3d_mask>0,(100,100,100),(255,255,255))
elif parcel_type == config.ParcelType.O:
color3d_mask = np.where(color3d_mask>0,(0,255,255),(255,255,255))
elif parcel_type == config.ParcelType.A:
color3d_mask = np.where(color3d_mask>0,(51,255,255),(255,255,255))
elif parcel_type == config.ParcelType.B:
color3d_mask = np.where(color3d_mask>0,(102,255,255),(255,255,255))
elif parcel_type == config.ParcelType.C:
color3d_mask = np.where(color3d_mask>0,(153,255,255),(255,255,255))
elif parcel_type == config.ParcelType.U:
color3d_mask = np.where(color3d_mask>0,(40,0,255),(255,255,255))
build3d_mask = np.zeros(self.img.shape, dtype=np.uint8)
build3d_mask[:,:,:] = building_mask[:,:,np.newaxis]
if self.building_masks is not None:
self.building_masks &= build3d_mask
self.img_last &= self.img_partitioned & build3d_mask
self.img_built &= self.img_partitioned & build3d_mask & color3d_mask
else:
self.building_masks = build3d_mask
self.img_last = self.img_partitioned & build3d_mask
self.img_built = self.img_partitioned & build3d_mask & color3d_mask
if config.WRITE_UNNECESSARY:
cv2.imwrite(f'outputs/final_map_{iteration}_{map_id}_{parcel_id}.bmp',self.img_built)
def draw_building_results(self):
cv2.imwrite(f'outputs/buildings_mask.bmp',np.bitwise_not(self.building_masks))
cv2.imwrite(f'outputs/final_map_location_finding.bmp',self.img_built)
def draw_collision(self):
trees_mask = cv2.imread('outputs/tree_mask.bmp')
fixed_facility_mask = cv2.imread('outputs/facility_mask.bmp')
roads_mask = cv2.imread('outputs/roads_mask.bmp')
collide_mask = roads_mask
if os.path.exists('outputs/buildings_mask.bmp'):
collide_mask |= cv2.imread('outputs/buildings_mask.bmp')
if os.path.exists('outputs/partitioning_mask.bmp'):
collide_mask |= cv2.imread('outputs/partitioning_mask.bmp')
# change here when building mask is av
collision3dmask = trees_mask | fixed_facility_mask
collision3dmask = collide_mask & collision3dmask
img = self.img_last.copy()
pixels = [100,50,100]*int(len(img[collision3dmask>0])/3)
img[collision3dmask>0] = pixels
cv2.imwrite(f'outputs/collision_map.bmp',img)