Part 1 — Visual Feature Detection for Autonomous Vehicle Video Streams

Part 1 of GPS Denied Navigation

This is Part 1 in a series on building software for GPS Denied Navigation in advanced aerospace robotics.

• Part 1 — Visual Feature Detection for Autonomous Vehicle Video Streams
• Part 2 — The Math Behind Optical Flow
• Part 3 — Lucas-Kanade Optical Flow

This is what we’re building in this post:

Go here to see a hq version: https://gyazo.com/2a01bc9f55bb007f50125c2bbd7aeb2d

In this example I used the first 15 seconds of the comma ai speedchallenge video and ran a Shi-Tomasi feature detector on it. This isn’t that useful, but demonstrates how you can extract features from video.

We use cv.goodFeaturesToTrack() and set some default parameters to get some good data out of it.

This is the first frame. Loop through the frames of the video, find the corners by turning the frame to grayscale then running it through opencv’s Shi-Tomasi algo.

Check out Part 2!