Building a Path Marker Detector

June 22, 2023
A bounding box generated from simulated path marker footage.

The Problem

Path markers, which are long, orange boards placed on the pool bottom during competition, point in the direction of the next task at the end of a completed one. This comes especially handy in environments like TRANSDEC, where darkness may obscure our submarine’s ability to navigate around the pool. Yet, path markers can only can be helpful if our submarine is able to detect them and recognize which direction they are pointing. We modeled this situation by developing Blender models of these path markers and generating virtual footage of what our submarine might see during competition as seen below.

Computer Vision to the Rescue

Now that we know what kind of footage we can expect to see from our submarine, we can apply computer vision approaches to detect the marker and move in the correct direction. Specifically, we take advantage of the distinct orange color of the marker to first differentiate it with the rest of the frame. Once we identify the rough orientation of the path marker, we find what direction to travel by mapping points on our detected marker to the internal map of the submarine. This allows submarine to add to its internal map and efficiently travel to the next challenge.