def img2binList(img, lenWidth, GRID_SIZE=50, verbose=0):
global DISTANCECOSTMAP
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, gray = cv2.threshold(gray, 112, 255, cv2.THRESH_BINARY_INV)
if verbose:
cv2.imshow("img", gray)
cv2.waitKey(0)
cnts = cv2.findContours(gray.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
cnts = imutils.grab_contours(cnts)
locs = []
height, width = gray.shape
tmp = np.zeros((height, width), np.uint8)
idxLargest = 0
areaLargest = 0
# loop over the contours
for (i, c) in enumerate(cnts):
# compute the bounding box of the contour, then use the
# bounding box coordinates to derive the aspect ratio
(x, y, w, h) = cv2.boundingRect(c)
if w * h > areaLargest:
idxLargest = i
areaLargest = w * h
cv2.rectangle(tmp, (x, y), (x + w, y + h), (255, 0, 0), 2)
if verbose:
# print("found largest contour outline")
cv2.imshow("img", tmp)
cv2.waitKey(0)
# print("cropping image as largest contour")
(x, y, w, h) = cv2.boundingRect(cnts[idxLargest])
gray = gray[y:y + h, x:x + w]
if verbose:
cv2.imshow("img", gray)
cv2.waitKey(0)
mapWidth = (int)(lenWidth // GRID_SIZE)
mapHeight = (int)((h / w) * lenWidth // GRID_SIZE)
print("the map will be created by the size: " + str(mapWidth) + " X " + str(mapHeight))
resized_gray = imutils.resize(gray, width=mapWidth) # resize the map for convolution
_, resized_gray = cv2.threshold(resized_gray, 1, 255, cv2.THRESH_BINARY)
if verbose:
cv2.imshow("img", resized_gray)
cv2.waitKey(0)
maze = convert2list(resized_gray)
my_maze = np.array(maze)
solution = pyfmm.march(my_maze == 1, batch_size=100000)[0] # NOTE : white area means walkable area
DISTANCECOSTMAP = solution
# cv2.destroyAllWindows()
return maze
def test_fmm_infpick(self):
# Given
my_mesh = np.zeros((5, 5), dtype=np.bool)
my_mesh[2, 2] = True
# When
exact = pyfmm.march(my_mesh, batch_size=np.inf)[0]
# Then
if self.do_plot:
plt.imshow(exact, interpolation='None')
plt.colorbar()
plt.show()
self.assertFalse(np.argwhere(np.isnan(exact)))
self.assertAlmostEqual(exact[2, 2], 0)
self.assertAlmostEqual(exact[0, 2], 2.0)
self.assertAlmostEqual(exact[2, 0], 2.0)
def test_fmm_npick(self):
# Given
my_mesh = np.zeros((5, 5), dtype=np.bool)
my_mesh[2, 2] = True
# When
solution = pyfmm.march(my_mesh, batch_size=5)[0]
# Then
if self.do_plot:
plt.imshow(solution, interpolation='None')
plt.colorbar()
plt.show()
self.assertFalse(np.any(np.isnan(solution)))
self.assertAlmostEqual(solution[2, 2], 0)
self.assertAlmostEqual(solution[0, 2], 2.0)
self.assertAlmostEqual(solution[2, 0], 2.0)
def test_fmm_L_shape(self):
# Given
my_mesh = np.zeros((200, 200), dtype=np.bool)
my_mesh[75:125, 75] = True
my_mesh[75, 75:125] = True
# When
solution = pyfmm.march(my_mesh, batch_size=100)[0]
#solution = pyfmm.pyfmm(my_mesh, n_min_pick_size=20) TODO: investigate why some values are undefined (inf or nan)
# Then
if self.do_plot:
plt.imshow(solution, interpolation='None')
plt.colorbar()
plt.show()
self.assertFalse(np.any(np.isinf(solution)))
self.assertFalse(np.any(np.isnan(solution)))
self.assertTrue(np.max(solution) < 160)
def test_fmm_race_to_middle(self):
# Given
my_mesh = np.zeros((20, 200), dtype=np.bool)
my_mesh[:, 0] = True
my_mesh[:, -1] = True
speed_map = np.ones((20, 200))
speed_map[:, 100:] = 2.0
# When
solution = pyfmm.march(my_mesh, speed=speed_map, batch_size=20)[0]
# Then
if self.do_plot:
plt.imshow(solution, interpolation='None')
plt.colorbar()
plt.show()
self.assertFalse(np.any(np.isinf(solution)))
self.assertFalse(np.any(np.isnan(solution)))
self.assertTrue(np.max(solution) < 75)
def test_fmm_plateau(self):
# Given
my_mesh = np.zeros((20, 200), dtype=np.bool)
my_mesh[:, 0] = True
speed_map = np.ones((20, 200))
speed_map[:, 100:] = 0.0
# When
solution, certain_values = pyfmm.march(my_mesh,
speed=speed_map,
batch_size=20)
# Then
if self.do_plot:
plt.imshow(solution, interpolation='None')
plt.colorbar()
plt.show()
self.assertLessEqual(np.sum(np.isinf(solution)), 2000)
self.assertLessEqual(np.sum(certain_values), 2000)
def test_fmm_known_values(self):
# Given
xx = np.outer(np.arange(0, 20, 1), np.ones((20, )))
yy = np.outer(np.ones((20, )), np.arange(0, 20, 1))
true_solution = np.abs(np.sqrt(np.square(xx) + np.square(yy)) - 10)
known_values = true_solution < 1.0
# When
computed_solution, _discard = pyfmm.march(known_values,
true_solution,
batch_size=1)
# Then
if self.do_plot:
plt.imshow(computed_solution - true_solution, interpolation='None')
plt.colorbar()
plt.title('test_fmm_known_values')
plt.show()
self.assertLess(np.max(np.abs(computed_solution - true_solution)), 0.7)
def test_fmm_circle(self):
# Given
n = 200
my_mesh = np.zeros((n, n), dtype=np.bool)
xx = np.array(100 + 50 * np.cos(np.linspace(0, 2 * np.pi, 100)),
dtype=np.int)
yy = np.array(100 + 50 * np.sin(np.linspace(0, 2 * np.pi, 100)),
dtype=np.int)
my_mesh[xx, yy] = True
# When
solution = pyfmm.march(my_mesh, batch_size=10)[0]
# Then
if self.do_plot:
plt.imshow(solution, interpolation='None')
plt.colorbar()
plt.show()
self.assertFalse(np.any(np.isinf(solution)))
self.assertAlmostEqual(solution[n / 2, n / 2], 50.0, places=-1)
self.assertAlmostEqual(solution[0, 0], 1.4142 * 100 - 50, places=-1)
import matplotlib.pyplot as plt
import pyfmm
my_image = plt.imread('irregular_boundary.png')
solution = pyfmm.march(my_image[:, :, 0] == 0, batch_size=10)[0]
plt.imshow(solution, interpolation='None')
plt.colorbar()
plt.title('Irregular boundary')
plt.show()
import numpy as np
import matplotlib.pyplot as plt
import pyfmm
# Double speed, 50/50
my_mesh = np.zeros((20, 200), dtype=np.bool)
my_mesh[:, 0] = True
my_mesh[:, -1] = True
speed_map = np.ones((20, 200))
speed_map[:, 100:] = 2.0
solution_a = pyfmm.march(my_mesh, speed=speed_map, batch_size=5)[0]
solution_b = pyfmm.march(my_mesh, speed=speed_map, batch_size=np.inf)[0]
plt.subplot(3, 2, 1)
plt.imshow(speed_map, interpolation='None')
plt.colorbar(orientation='horizontal')
plt.title('Speed map (50/50)')
plt.subplot(3, 2, 3)
plt.imshow(solution_a, interpolation='None')
plt.colorbar(orientation='horizontal')
plt.title('Accurate solution (50/50)')
plt.subplot(3, 2, 5)
plt.imshow(solution_b, interpolation='None')
plt.colorbar(orientation='horizontal')
plt.title('Inaccurate solution (50/50)')
# 10x speed, 90/10
import pyfmm
import matplotlib.pyplot as plt
import numpy as np
from time import perf_counter
# Define a boundary resembling a circle
n = 300
my_mesh = np.zeros((n, n), dtype=np.bool)
xx = np.array(100 + 50 * np.cos(np.linspace(0, 2 * np.pi, 100)), dtype=np.int)
yy = np.array(100 + 50 * np.sin(np.linspace(0, 2 * np.pi, 100)), dtype=np.int)
my_mesh[xx, yy] = True
# Compute solution
tmp = perf_counter()
solution_n1 = pyfmm.march(my_mesh, batch_size=1)
t1 = perf_counter() - tmp
tmp = perf_counter()
solution_n20 = pyfmm.march(my_mesh, batch_size=20)
t2 = perf_counter() - tmp
tmp = perf_counter()
solution_n100 = pyfmm.march(my_mesh, batch_size=100)
t3 = perf_counter() - tmp
tmp = perf_counter()
solution_ninf = pyfmm.march(my_mesh, batch_size=np.inf)
t4 = perf_counter() - tmp
print("Timings (s):")
print("\tbatch_size = 1: %2.3f" % t1)
print("\tbatch_size = 20: %2.3f" % t2)
print("\tbatch_size = 100: %2.3f" % t3)
print("\tbatch_size = inf: %2.3f" % t4)
