Uploaded on GitHub Repository: https://github.com/yeongseok94/CarND-LaneLines-P1.git
Forked from Self-Driving Car Nanodegree Project 1: https://github.com/udacity/CarND-LaneLines-P1.git
Finding Lane Lines on the Road
The goals / steps of this project are the following:
- Make a pipeline that finds lane lines on the road
- Reflect on your work in a written report
- "helperfunction.py": Copied version on helper functions described in P1.ipynb. Not used in running of the code.
- "lanedetection_test.py": Test code used for detailed test.
- "lanedetection_video.py": Main code for lane detection on the test videos.
- "processimage.py": Necessary classes and functions for lane detection. Especially, "LaneDetection.process_image()" shows the overall pipeline.
Overall pipline can be easily found in "LaneDetection.process_image()".
Step 1. Grayscale Image
First, I grayscaled the image for further processing. Here, I used openCV "cvtColor" function.
Step 2. Obtain Gaussian Blurred Image
Next, I removed noises with Gaussian filter of kernel size 5. Here, I used openCV "GaussianBlur" function.
Step 3. Obtain Edges with Canny Transform
Next, I obtained edges with Canny transform. Here, I used openCV "Canny" function. The low and high thresholds are set as 50 and 150.
Step 4. Mask onto Canny Transformed Image
Next, I masked the possible lane region onto Canny transformed image. Here, I used openCV "fillPoly" function. The region of interest is trapezoidal area that covers about lower quarter of the image slightly above from the bottom of the image. Note that the bottom of the region is not equal to the bottom of the image in order to reject fron area of the car. Vertices of the trapezoidal area are defined as relative value proportional to image size.
Step 5. Line Candidates by Hough Transform
Next, I obtained line candidates by Hough transform. Here, I used openCV "HoughLinesP" function. I selected distance resolution as 1px, angular resolution as 1deg, minimum number of votes as 1, minimum line length as 40px, and maximum line gap as 20px.
Step 6. Selection of Left and Right Lines of the Lane
Final procedure includes deciding the lane lines with obtained line candidates.
First, I categorized the line candidates into left line candidates and right line candidates with slopes of the lines. Right line candidates should have positive slope values, and left line candidates should have negative slope values.
Next, I narrowed down each line candidates by checking where on the bottom of the image the extrapolated line segments go through. Proper right line candidates should go through right half of the bottom of the image, and left line candidates should go through left half of the bottom of the image.
Next, I again narrowed down each line candidates by checking whether the slope of the line is inside ±5deg with respect to the median value among them. Here I simply selected median value as a priori estimate of the slope, but it can be some other value. This procedure is to reject unexpected line expectation.
Finally, I selected one single right lane line and left lane line by averaging the line parameters of the line candidates and visualized it on the image.
The output lane line has oscillation as time goes by. This is because the proposed pipeline only includes the processing within the frame, and does not consider the time-varying effect of the lane line.
Also, there is such cases that no line candidates left, therefore no output is generated.
Additionally, this code does not include adjusting thresholds on canny transform, so the output is highly sensitive to the overall brightness around the lane line area.
The oscillation of the output lane line can be improved by introducing proper lowpass filter or time averaging, Kalman Filter, etc. This is possible since lane lines always moves quite slow, i.e. there is no such cases that lane line jumps across hundreds of pixels in one frame.
Also, if there is no line candidates left, we can just preserve the previous lane line output or possibly extrapolate line variables with respect to time.
Additional considerations on brightness adjustment around the lane line area and consideration on the colors of the pixels can improve the performance.