Monday May 12, 4:00PM
The problem of recognizing hand gestures in real-time from a video feed has been thoroughly studied in the field of computer vision, but methods for real-time hand gesture recognition on low performance machines are scarce. In this work, we investigate several different algorithms for hand gesture recognition on a Raspberry Pi, a cheap computer with a low performance CPU and a relatively high performance GPU. We develop two different methods for hand gesture recognition on the Pi: image histogram comparison and contour comparison based on sample gesture images. For each of these two main approaches, we first implement a simple algorithm and a more sophisticated version. We then compare the speed and accuracy of all our algorithms to pick the approach that yields the best accuracy while still achieving real-time analysis.
Refreshments will be served.