Device-free locators allow the user to interact with a system without the burden of being physically in contact with some input device or without being connected to the system with cables. This thesis presents a device-free locator that uses computer vision techniques to recognize and track the user's hand. The system described herein uses a video camera to capture live video images of the user, which are segmented and processed to extract features that can be used to locate the user's hand within the image. Two types of features, namely moment based invariants and Fourier descriptors, are compared experimentally. An important property of both these techniques is that they allow the recognition of hand-shapes regardless of affine transformation, e.g. rotation within the plane or scale changes. A neural network is used to classify the extracted features as belonging to one of several hand signals, which can be used in the locator system as 'button clicks' or mode indicators. The Siltrack system described herein illustrates that the above techniques can be implemented in real-time on standard hardware.
Dissertation (MSc (Computer Science))--University of Pretoria, 2007.