Had an idea to add to this robot I have been working on for school. I want to know the feasibility of using a recycled camera, to get basic vision for my robot to run TensorFlow/TensorFlow Lite and determine if an audience member's "pose" is positive or negative.
I ripped the electronics out of a knockoff GoPro, and kept the USB interface intact. I got a micro-USB breakout to make the connection between the two. I've only been in the Arduino ecosystem for a few years so I don't know if it's possible, but is it possible to transfer image data through serial?
Results wise, I don't care if it is high or low definition, color or mono. Just enough to identify a human.
There are only a limited number of Arduino-compatible boards capable of running Tensor Flow Lite. Check the list on the Tensor Flow Lite site before you pick an incompatible model.
Almost never feasible. The ESP32-CAM is the way to give vision to your robot.
determine if an audience member's "pose" is positive or negative.
That would be a project for a very good camera feeding into a large ML program running in the cloud. If that has not already been done (and the service available upon subscription) it will be soon.
I am only saying this for you to ask yourself if there might be a better use for the cam, or a different way to analyze an audience. I am not making insinuations or assumptions in thisbquestion; "When is it ever a good idea to take pictures of children without their parents' consent?"
Hey John, I'm actually using an ESP32S2. I'm just more comfortable with the Arduino-semantics C++. My specific S2 dev module doesn't support TFL. Espressif has ESP NOW, and brief research shows I can handle it within Arduno too, so I may try that route.
This is part of a larger exhibition in an academic setting. The visual data just gets erased and I have kept data ethics in mind, no worries. There's an intentionality behind choosing a camera for audience observation, but it's not a high-priority feature.