This week, our focus was to develop possible concepts for the prototype design. The concept generation process was split up between hardware and software. See the table below or a summary of our concept ideas. 

 Software

A possible software implementation option is Google’s open-source gesture recognition package with TensorFlow Lite and MediaPipe. The package recognizes a variety of one-handed gestures, and the test recognition results are accurate. This package is not the only option; the team will consider more open source packages to aid development. Another option is the gesture recognition package from Tencent, which seems to be complete and supports various systems and application environments.  

 Hardware

Three main options exist for the overall hardware architecture. All-in-one solutions like the Azure Kinect DK already have all the necessary sensing equipment (and more) built into one set-top box. This is the most expensive and most powerful option.  Simplified solutions like the OmniVision image sensors have some of the necessary sensing hardware and are much less expensive. In this case, the prototype would likely require additional hardware. Creating the entire system from individual hardware components would be the most complex option to design (though likely the cheapest to produce). Many other hardware-related decisions will stem from this overall hardware architecture concept. In the individual component case, concept finalization will require deciding on the types of sensors and microcontroller(s). Essentially, this method would be attempting to recreate an all-in-one solution like the Azure Kinect DK but with only the necessary components.