Autonomous Vision & Robot Control

This one went through two versions. The first runs on recorded road video — it picks out the lane lines, finds the middle of the lane, guesses if the road is turning left, right, or going straight, and draws all of that back on the video.

The second version does it on a live camera feed from the robot and adds a control panel (Tkinter) that sends movement commands to the robot over HTTP.

First I crop each frame, pick out the yellow lane lines by color, and warp it so it's like looking down at the road from above. Then it goes grayscale, Canny finds the edges, and a Hough transform turns those edges into line segments.

Next I toss out lines that aren't close to vertical and split the rest into left and right lanes based on where they are. Averaging each side gives me the lane lines plus a center line, and where that center line sits decides the turn call. Last step, the overlay gets warped back to the normal view and shown in the app.

The live robot version runs a simpler take on this on the camera feed: grayscale, blur, Canny, Hough, and a slope filter.