Gesture Recognition Based on Computer Vision on a Standalone System

Abstract

Our project uses computer vision methods gesture recognition in which a camera interfaced to a system captures real time images and after further processing able to recognize the gesture shown to be interpreted. Our project mainly aims at hand gestures and after extracting information we try to produce it as an audio or in some visual form. We have used adaptive background subtraction with Haar classifiers to implement segmentation then we used convex hull and convex defects along with other feature extraction algorithms to interpret the gesture. First, this is implemented on a PC or laptop and then to produce a standalone system, we have to perform all this steps on a system which is dedicated to perform only the given specified task. For this we have chosen Beaglebone Black as a platform to implement our idea. The development comes with ARM Cortex A8 processor supported by NEON processor for video and image processing. It works on a clock frequency of maximum 1 GHz. It is 32 bit processor but it can be used in thumb mode i.e. it can work in 16 bit mode. This board supports Ubuntu, Android with some modification. Our first task is to interface a camera to the board so that it can capture images and store those as matrices followed by our steps to modify the installed Operating System to our purpose and implement all the above processes so that we can come up with a system which can perform gesture recognition.