At the University of Wisconsin - Stout’s Research in Robotics for Assistive Technology REU, I had the privilege to spend a summer in Menomonie, WI, researching solutions in wheelchair technology. Current navigation systems use lidar or GPS for tracking, neither of which works well for fine-tuning sidewalk trajectory. I chose my handy Raspberry Pi to tackle the problem. Using a webcam and computer vision, I put together an algorithm that detected the center of a sidewalk and encouraged the user to follow it. The trick is to use Lab color space. You can find the “distance” between colors by using Euclidean distance. The algorithm is as follows:

• Sample colors directly in front of wheelchair
• Remove colors a certain color distance away
• Morphologically dilate then erode to get a nice shape of the sidewalk
• Find the centroid of the remaining shape
• Resist wheelchair direction proportionally

Github Repo