region_masking

region_masking.py

# Region masking is a technique used to run a image processing or computer vision task within a specific region of interest
# Step 1. Build a selector for the region
# Step 2. Build a selector for the pixels of interest
# Step 3. AND the two selectors. It's that simple.

from matplotlib import image
from matplotlib import pyplot
import numpy

view_from_windshield = image.imread('test.jpg')


# Building the region selector
canvas = view_from_windshield.shape
#pyplot.imshow(view_from_windshield)
#pyplot.show()

# Set a triangle region
# We need left, right and apex co-ordinates (x,y) (0,0) is top left
print(canvas)
# x,y
left = [0,540]
right = [960,540]
apex = [480,300]
# y = ax+b
# solve for a and b
# polyfit((x1,x2,x3..),(y1,y2,y3...), degree_of_ploynomial_to_fit)
# for straight line polyfit((x1,x2),(y1,y2),1) output a tuple of two parameters
left_line = numpy.polyfit((left[0], apex[0]),(left[1], apex[1]),1) 
right_line = numpy.polyfit((right[0], apex[0]),(right[1], apex[1]),1)
base_line = numpy.polyfit((left[0], right[0]),(left[1], right[1]),1)
# create co-ordinates using meshgrid
# x,y = numpy.meshgrid([0,1,...539],[0,1,...959]) for image 960 x 540
# x,y is a matrix of co-ordinate values (x,y) denotes a point.
# arange(n,m) is n included to m not included so it takes care of our range 0, 539 instead of 0, 540.
x,y = numpy.meshgrid(numpy.arange(0,canvas[1]), numpy.arange(0,canvas[0]))
# The region of interest is (x>0&y>left_line[0].x+left_line[1]) but x>0 is always true
# similarily (y>right_line[0].x+right_line[1]) and y<base_line[0].x+base_line[1]
region_selector = (y>(left_line[0]*x+left_line[1]))&(y>(right_line[0]*x+right_line[1]))&(y<(base_line[0]*x+base_line[1]))


# Building the color selector
threshold = [200,200,200]
working_copy = numpy.copy(view_from_windshield)
working_copy2 = numpy.copy(view_from_windshield)
pixel_selector = (working_copy[:,:,0] < threshold[0])|(working_copy[:,:,1] < threshold[1])|(working_copy[:,:,2] < threshold[2])

working_copy2[~pixel_selector & region_selector] = [255,0,0]

pyplot.imshow(working_copy2)
pyplot.show()

But, as it happens, lane lines are not always the same color, and even lines of the same color under different lighting conditions (day, night, etc) may fail to be detected by our simple color selection.

What we need is to take our algorithm to the next level to detect lines of any color using sophisticated computer vision methods.