!apt update && apt install -y python3-opencv !pip install opencv-python

import cv2 import numpy as np import os import subprocess import glob from matplotlib import pyplot as plt INPUT_FOLDER = "./input/" OUTPUT_FOLDER = "./output/" TEST_IMAGE = "test_image.jpg" TEST_VIDEO = "test_video.mov" DEBUG = False def plot_image(image, title): plt.imshow(image, cmap=plt.cm.gray) plt.title(title) plt.show()

original_image = cv2.imread(INPUT_FOLDER + TEST_IMAGE) plot_image(original_image, 'Original image')

test_image = cv2.imread(INPUT_FOLDER + TEST_IMAGE) test_image = cv2.cvtColor(test_image, cv2.COLOR_BGR2GRAY) plot_image(test_image, 'Grayscale image')

test_image = cv2.GaussianBlur(test_image, (3,3),0) plot_image(test_image, 'Blur image')

test_image = cv2.Canny(test_image, 100, 150) plot_image(test_image, 'Canny image')

def roi(image): bottom_padding = 100 # Front bumper compensation height = image.shape[0] width = image.shape[1] # FYI, below values are highly dependant on the camera calibration i.e what part of the road is actually being captured bottom_left = [0, height-bottom_padding] bottom_right = [width, height-bottom_padding] top_right = [width*1/3, height*1/3] top_left = [width*2/3, height*1/3] vertices = [np.array([bottom_left, bottom_right, top_left, top_right], dtype=np.int32)] mask = np.zeros_like(image) cv2.fillPoly(mask, vertices, 255) plot_image(mask, "mask") masked_image = cv2.bitwise_and(image, mask) return masked_image

roi_image = roi(test_image)

plot_image(roi_image, "Masked image")

def averaged_lines(image, lines): right_lines = [] left_lines = [] for x1,y1,x2,y2 in lines[:, 0]: parameters = np.polyfit((x1, x2), (y1, y2), 1) slope = parameters[0] intercept = parameters[1] if slope >= 0: right_lines.append([slope, intercept]) else: left_lines.append([slope, intercept]) def merge_lines(image, lines): if len(lines) > 0: slope, intercept = np.average(lines, axis=0) y1 = image.shape[0] y2 = int(y1*(1/2)) x1 = int((y1 - intercept)/slope) x2 = int((y2 - intercept)/slope) return np.array([x1, y1, x2, y2]) left = merge_lines(image, left_lines) right = merge_lines(image, right_lines) return left, right def hough_lines(image, rho, theta, threshold, min_line_len, max_line_gap): lines_image = np.zeros((image.shape[0], image.shape[1], 3), dtype=np.uint8) lines = cv2.HoughLinesP(image, rho, theta, threshold, np.array([]), minLineLength=min_line_len, maxLineGap=max_line_gap) if lines is not None: lines = averaged_lines(image, lines) for line in lines: if line is not None: x1,y1,x2,y2 = line cv2.line(lines_image, (x1, y1), (x2, y2), (0, 0, 255), 20) plot_image(lines_image, "lines") return lines_image

hough_lines_image = hough_lines(roi_image, 0.9, np.pi/180, 100, 100, 50)

def combine_images(image, initial_image, α=0.9, β=1.0, λ=0.0): combined_image = cv2.addWeighted(initial_image, α, image, β, λ) plot_image(combined_image, "combined") return combined_image

final_image = combine_images(hough_lines_image, original_image)