DSP Based Gesture Recognition

Abstract

 Gesture Recognition lies at the core of augmented realitybased devices. This project tries to coalesce two disparate aspects ofengineering which are image processing and embedded systems. Theproject aims at developing and algorithm for recognizing staticgestures using deterministic techniques and implementing thesealgorithms on an embedded platform. The first part of the projectdeals with the development of the algorithm.The underlying principle of the gesture recognition is singular valuedecomposition and normalized distances. These concepts are used incomparison and classification of query gestures with databasegestures. Principal Component Analysis is used for dimensionalityreduction and ease of calculations. Image processing techniques areused prior to computation to achieve background, illumination andscale invariant images. The database comprises of such preprocessedimages stored with varying orientations. The other part of the projectdeals with setup of an embedded platform for the compilation andimplementation of the algorithms. A DSP based board is configuredto interface a camera and input devices. Software setup is done tosupport graphics and OpenCV. Compilation of the programs writtenfrom the algorithms is carried out on the board. Thus, the projectaims at standalone implementation of gesture recognition which setsthe foundation for portable augmented reality.