Finding_Lane_lines

1.Pipeline

(1) Transform the image into grayscale

use cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

(3) Blur the image using a 7*7 kernel

blur = cv2.GaussianBlur(gray, (7, 7), 0)

(4) Use canny edge detection to find all edges in the image

(5) Crop the image with a triangle

the triangle size is: [(int(0.08 * width), height), (int(0.92 * width), height), (int(0.5 * width), int(height * 0.3))]

(6) Using Hough to extract straight lines

Here are my settings: minLineLength = 60 maxLineGap = 10 lines = cv2.HoughLinesP(lineImg, 1, np.pi / 120, 100, minLineLength, maxLineGap) These are the lines found by HoughLinesP:

(7) separate left and right lines, calculate the slope, and draw green lines with the points found by getCoor()

2.Shortcomings and future improvement

(1). The green line disappears periodically in the output video.

This is because in a frame like below, the right side lane less clear comparing the first frame I showed in this README. The first right white line we can se is farther way from the camera. while in a frame like below, where the first right side line is closer to the camera, then the straight line is longer and easier to detect on the picture improvements: As we can see from step 4, edge detection has destinguished the edge. So the algorithm for Hough and calculating lines should be improved. Also, I could extends the line detected from last frame and carry over to make the video look more smooth.

(2). This model assumes road being straight

If the road bends, the triangle in this model will no longer be able to detect appropriate lane area. To imporve this, it needs to be broaden, maybe with a trapazoid
if the road bends at a large angle, then it won't be detected as a straight line by Hough

(3) Houghline not detecting best result

Hough Transform is tricky to get right with its parameters. I am not sure I got the best settings.

an-bluecat / finding_lane_lines Goto Github PK