Machine Vision Experiments

To provide additional processing oomph we have taken the plunge to upgrade to a much more powerful Pi 4b.  Surprisingly, it does not seem to draw anymore power than the old 3b.  It could be we are not really pushing it that hard yet.

We tried to upgrade to the latest 'Bookworm' OS to take advantage of 64bit processing, but it has proven much harder than originally thought.  They main problem is the backward compatibility of the 'Approxeng' module we use to connect to a gamepad controller.  For now we are sticking with the 'Buster' OS.  This may cause us problems later on if we require updated dependencies.

The video shows our attempts at 'machine vision'.  We are concentrating on the 'Escape Route' challenge first which involves navigating a maze with moveable 'coloured' walls.  Our initial approach simply used the ultrasonic distance sensors and a simple 'turn right' maze solving algorithm to navigate the maze, but we think 'machine vision' and the ability to recognise colour of the walls will improve performance.

The video shows our initial 'machine vision' algo, utilising openCV, works quite well at picking out the yellow cube.