Machine Vision Bubble Detection

Setup an automated system to watch for the presence of bubbles in a low cost bio-reactor made by a student for their PhD dissertation. Part of this project also involved 3D printing the walls of the bio-reactor out of polypropylene such that the reactor could be put in an autoclave for a full cleaning for reuse between experiments. The machine vision project involved only low powered edge devices. Namely a raspberry pi at the edge which collects camera images from an infrared camera and a RGB camera and sends a low latency stream via an RTSP stream to a Nvidia Jetson for the machine vision application. This involved setting up both platforms such that the RTSP stream would be accepted by the Jetson which was then used to train a convolutional neural network to pick up bubbles. The CNN was then implemented in python code to send a groupme message to the student if their aeration were to fall below an acceptable level.