The goal of this project was to create a system that utilized closed loop control in order to map some sensor reading to some actuator output.  To achieve this goal, we designed a system that would balance a ping pong ball on a clear sheet of acrylic.  The camera captures images of the acrylic sheet and ping pong ball.  These images are converted to HSV color space and thresholded to detect green pixels.  Morphological operations are performed to reduce noise in the binary image.  The center of the largest group of green pixels is used as our ping pong ball coordinates.  These coordinates are then transformed to a coordinate system in which the center of the acrylic is (0, 0) and the axes are rotated to be aligned with the diagonal of the acrylic sheet.  This simplifies the mapping from sensor input to motor output, as the motors raise each corner of the acrylic.  In order to ensure that the motors move in such a way that keeps the acrylic planar, we move the motors in opposite corners equal yet opposite amounts.  

Our first working prototype was able to balance a ping pong ball for approximately 17 seconds on average before failure.  This is likely due to issues with our Raspberry Pi camera module, which was not able to be recognized by our Raspberry Pi.  As a result, we implemented a workaround in which we used a smartphone to take images and upload these images to a URL.  We would then use wget to download each image, which would then be sent to our image processing file.  As a result, our frame rate was only about 5 fps.  Future versions will use a Raspberry Pi camera, drastically improving frame rate.  Additionally, the use of a wide angle lens will allow the acrylic to be brought much closer to the base while still allowing the entire acrylic sheet to be captured.  This will reduce tremors caused by frequent updates in motor duty cycle.